Ship intelligent power consumption method and system based on neural network

Abstract

The present invention discloses a method and system for intelligent electricity consumption of ships based on neural networks, including establishing a multi-season and multi-period prediction model, outputting predicted values ​​of ship power consumption according to the neural network; dividing charging probabilities of different time periods into multiple time scales by Monte Carlo method and Newton method, and obtaining charging probabilities of ships at different time scales; the present invention can calculate the charging probabilities of different time periods and formulate specific charging strategies by adopting Monte Carlo simulation and Newton method technology, so as to solve the problem of insufficient power supply or overcharging caused by the imbalance between power supply and demand of ships, and can effectively predict and dispatch the charging demand of ships by simulating behaviors at different times and geographical locations and using Newton method to divide charging probabilities of multiple time scales, so as to optimize the charging behaviors of ships, ensure the stability and efficiency of power supply, and also reduce energy costs.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent power management, and in particular to a method and system for intelligent power consumption of ships based on a neural network. Background Art

[0002] With the advancement of ship technology, modern ships are gradually developing towards automation and intelligence. In particular, in the power management system during navigation, accurate power consumption prediction and optimization have become the key. However, traditional power management methods are often unable to cope with complex ship operating data, especially in the context of multi-season, multi-time, and multi-geographical changes. Due to the large fluctuations in ship power demand under different environmental conditions, the existing system is difficult to capture these changes, resulting in an imbalance between power supply and demand, causing insufficient power supply or overcharging. Summary of the invention

[0003] The purpose of the present invention is to provide a method and system for intelligent power consumption of ships based on neural networks, so as to solve the problem of insufficient power supply or overcharging caused by the imbalance between power supply and demand of ships, improve energy utilization and reduce energy consumption costs.

[0004] The purpose of the present invention can be achieved by the following technical solutions:

[0005] In the first aspect, the present application provides a method for intelligent power consumption of ships based on a neural network, the steps of which include:

[0006] Receiving ship operating condition data from a data source, establishing a multi-season and multi-period prediction model, transmitting the received information as an input vector to a neural network of the multi-season and multi-period prediction model; and outputting a predicted value of ship power consumption according to the neural network;

[0007] According to the predicted value, the spatiotemporal data is extracted by Monte Carlo method, the behavior of the ship at different times and geographical locations is simulated, and the probability density function of the ship at different travel and stay times is calculated. Then, the Newton method is used to divide the charging probability of different time periods into multiple time scales, and the charging probability of the ship at different time scales is obtained;

[0008] Specifically, the spatiotemporal data is extracted through the Monte Carlo method according to the predicted value, and the behavior of the ship at different times and geographical locations is simulated to calculate the probability density function of the ship at different travel and stay times, including:

[0009] Set the value range of the time variable T and define the position of the ship and power consumption as a function of time and treated as a random variable;

[0010] According to T, and The probability distribution of the random sampling is used to generate simulation samples. Each simulation sample represents a ship behavior path. The more samples are taken, the more accurate the simulation results are.

[0011] By analyzing the simulated samples, the probability distribution of ships at different times and locations is obtained, and the path probability is calculated. as the frequency of occurrence of the path;

[0012] For continuous data, use the limit definition to calculate the vessel dwell time and ship travel time The probability density function of

[0013] The Newton method is used to divide the charging probability of different time periods into multiple time scales, and the charging probability of the ship at different time scales is obtained, including:

[0014] Set the first objective function , represents the expected value of the total power consumption of the ship at different time scales, and is a function of the charging probability P, where the first objective function is expressed as:

[0015] ;

[0016] in, It is the time period The weight of ) is in the time period The expected value of power consumption under It is in the time period 2 corresponding charging probability;

[0017] Use Newton's method to find the optimal charging probability so that the first objective function The optimal, specific Newton method iteration is expressed as: ,in is the charging probability at the kth iteration, The first objective function is The first-order derivative at , is the second-order derivative.

[0018] Furthermore, a multi-season and multi-period prediction model is constructed by integrating CNN and RNN models, including:

[0019] Connect the image feature vector extracted by CNN and the time series feature vector output by RNN to fuse them into a feature vector;

[0020] Establish a fully connected layer to process the fused feature vector through nonlinear transformation, and input the processed feature vector into the multi-season and multi-period prediction model;

[0021] The predicted value of the ship's power consumption is output by calculating the forward propagation.

[0022] Furthermore, the prediction of the ship power consumption is calculated by the forward propagation output through the first neural network. Specifically, the fused feature vector is input into the fully connected layer, and the output of each fully connected layer is calculated through forward propagation, wherein the forward propagation calculation is expressed as: ;

[0023] in, It is The output vector of the layer, for layer 1 ; It is The weight matrix of the layer is expressed as ; It is The bias vector of the layer, dimension is , It is The activation function of the layer, is the predicted value of the neural network.

[0024] During the forward propagation process, the output of each layer depends on the output of the previous layer until the final output layer generates a predicted value.

[0025] Furthermore, it also includes establishing a multi-objective model according to the charging probability, solving the first objective function by improving the non-dominated sorting genetic algorithm, and obtaining the power consumption strategy in different time periods;

[0026] Furthermore, the first objective function is solved by improving the non-dominated sorting genetic algorithm to obtain the power consumption strategies for different time periods, including:

[0027] The initial population is randomly generated, and each individual represents a potential electricity usage strategy, including the probability of charging in different time periods;

[0028] Perform non-dominated sorting on the current population, and distinguish individuals at each non-dominated level by calculating the dominance number and dominated set of each individual;

[0029] The population diversity is maintained by calculating the crowding distance of each individual, and individuals are selected based on their non-dominated rank and crowding degree;

[0030] Combine selection, crossover and mutation operations to generate a new offspring population, merge the parent population with the offspring population, perform non-dominated sorting and crowding calculation again, and select a new parent population based on the elite strategy;

[0031] Generate a set of Pareto optimal solutions, representing the optimal electricity consumption strategies in different time periods.

[0032] Furthermore, it also includes using a deep deterministic policy gradient algorithm to control ship power according to the power consumption strategy.

[0033] Furthermore, the deep deterministic policy gradient algorithm is used to control the ship power according to the power consumption strategy, including:

[0034] The predicted value is combined with the current ship status information and input into the DDPG algorithm as the current status;

[0035] Generate control instructions through the power strategy network;

[0036] The effectiveness of the generated control instructions is evaluated through the value network, and the parameters of the strategy network are adjusted through SAC to optimize the control decision;

[0037] The control instructions generated by the DDPG algorithm are sent to the ship's power management system to perform power regulation operations.

[0038] Furthermore, the Bellman equation is used to calculate the action value to evaluate the effectiveness of the current control instruction. The action value is expressed as: ;

[0039] in, Indicates taking action at time t The expected return is parameterized by the value network as , Indicates the current state and actions The immediate reward (or reward function) under is the feedback obtained immediately after the agent performs an action. represents the discount factor, which is used to measure the value of future rewards relative to immediate rewards, Indicates that at the next time point t+1 Take Action expected return; when the calculated action value is higher than the average level of the environment, it means that the control instruction of the action in the current state is valid; otherwise, the control instruction of the action in the current state is invalid.

[0040] Furthermore, the parameters of the policy network are adjusted through SAC, which is specifically expressed as:

[0041] ;

[0042] in, Represents the gradient symbol, for the policy network parameters The gradient operator is used to find the second objective function The rising direction under the current parameters; represents the expected inner product, which is for the state The expectation of the state is sampled from the experience buffer D. represents the entropy regularization term, is a hyperparameter that controls the weight of the entropy term, is in state Under the policy network Select Action The logarithmic probability of .

[0043] On the other hand, a neural network-based ship intelligent power consumption system is also provided, which is applied in a neural network-based ship intelligent power consumption method, including a data receiving and prediction module, a Monte Carlo simulation and Newton method optimization module, an intelligent power consumption strategy optimization module and a power regulation and control instruction execution module.

[0044] The data receiving and prediction module is used to receive ship operating condition data from a data source, establish a multi-season and multi-period prediction model, receive information as an input vector, and output a predicted value of ship power consumption using a neural network;

[0045] The Monte Carlo simulation and Newton method optimization module uses the Monte Carlo method to extract spatiotemporal data, simulate ship behavior, calculate the probability density function of ships under different travel and stay times, and use the Newton method to divide the multi-time scale charging probability in different time periods;

[0046] The intelligent power consumption strategy optimization module establishes a multi-objective model based on the charging probability, solves the second objective function by improving the non-dominated sorting genetic algorithm, generates a Pareto optimal solution, and obtains the optimal power consumption strategy for different time periods;

[0047] The power regulation and control instruction execution module combines the power consumption strategy and the deep deterministic policy gradient algorithm to perform power regulation, uses the SAC algorithm to adjust the policy network parameters, optimizes the control decision, and sends the generated control instructions to the ship's power management system for execution.

[0048] The beneficial effects of the present invention are:

[0049] A multi-season and multi-period prediction model is established using a neural network model to learn and predict ship operating data, and a hybrid model combining convolutional neural network (CNN) and recurrent neural network (RNN) is used to process and analyze data. This can solve the problem that traditional prediction methods are difficult to accurately capture these complex nonlinear relationships. By combining the hybrid structure of CNN and RNN, time series data and image data can be processed simultaneously, and the spatiotemporal features in ship operating data can be effectively extracted, which improves the prediction accuracy of ship power consumption, reduces energy waste and optimizes energy distribution.

[0050] By adopting technologies such as Monte Carlo simulation and Newton method, the charging probability in different time periods can be calculated, and specific charging strategies can be formulated to solve the problem of insufficient power supply or overcharging caused by the imbalance between the power supply and demand of ships. By simulating the behavior at different times and geographical locations, and using Newton method to divide the charging probability at multiple time scales, the charging demand of ships can be effectively predicted and scheduled, and the charging behavior of ships can be optimized, ensuring the stability and efficiency of power supply, while also reducing energy costs;

[0051] By combining the deep deterministic policy gradient (DDPG) algorithm and the soft actor-critic (SAC) to implement the control strategy, it is possible to adjust the power usage strategy according to real-time data and dynamically optimize the equipment operation status, solving the problems of power demand fluctuations and low energy efficiency caused by the changing environment faced by ships during navigation. At the same time, it also enhances the adaptability and robustness of the ship's power system, improves energy efficiency and reduces energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] For better understanding and implementation, the technical solution of the present application is described in detail below with reference to the accompanying drawings.

[0053] Figure 1 A schematic diagram of a process flow of a method for intelligent electricity use of a ship based on a neural network provided in Example 1 of the present application;

[0054] Figure 2 A schematic diagram of a process for constructing a multi-season and multi-period prediction model for a ship intelligent electricity consumption method based on a neural network provided in Example 1 of the present application;

[0055] Figure 3 A schematic diagram of a flow chart of obtaining power consumption strategies in different time periods for the method for intelligent power consumption of ships based on a neural network provided in Example 1 of the present application;

[0056] Figure 4 A schematic diagram of the process of using a deep deterministic policy gradient algorithm to control ship power in the neural network-based ship intelligent power consumption method provided in Example 1 of the present application. DETAILED DESCRIPTION

[0057] In order to further explain the technical means and effects taken by the present invention to achieve the predetermined invention purpose, exemplary embodiments will be described in detail here, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are only examples of methods and systems consistent with some aspects of the present application as detailed in the attached claims.

[0058] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0059] The specific implementation methods, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0060] Example 1

[0061] See also Figure 1-Figure 4 , this embodiment provides a method and system for intelligent power consumption of ships based on neural networks, which solves the problem of insufficient power supply or overcharging caused by the imbalance between power supply and demand of ships, improves energy utilization and reduces energy consumption costs.

[0062] The present invention provides a method for intelligent power consumption of ships based on a neural network, comprising the following steps:

[0063] S1. Receive ship operating condition data from a data source, establish a multi-season and multi-period prediction model, and transmit the received information as an input vector to a neural network of the multi-season and multi-period prediction model; output a predicted value of ship power consumption according to the neural network;

[0064] Specific data affecting ship operating conditions include various meteorological environmental conditions, ship power load, ship navigation status, ship equipment status information, etc.

[0065] It needs to be explained that ship power consumption has obvious time series characteristics, and power load and equipment status change over time. RNN, especially its variants such as long short-term memory network (LSTM), can process time series data well and capture dynamic characteristics in time series. Ship navigation status and various meteorological environmental conditions involve a variety of complex patterns and nonlinear relationships. CNN shows strong feature extraction capabilities in image and signal processing, which helps to extract useful features from the raw data of ship status. Therefore, choosing a hybrid model that combines CNN and RNN can take advantage of the advantages of both. CNN is responsible for extracting features, while RNN processes time series data. The combination of the two can improve the accuracy and robustness of predictions.

[0066] Furthermore, the marine working condition data received from the data source is preprocessed, and the preprocessed image data is processed using the CNN model and the RNN model, wherein the processing by the CNN model specifically includes: extracting deep features of the image through the convolution layer and the pooling layer; converting the output of the convolution layer into a feature vector of fixed length through the fully connected layer; processing by the RNN model includes: processing time series data through the RNN model to capture time features; outputting the hidden state of the last time step of the RNN model as a time series feature vector. Specifically, before building a multi-season and multi-period prediction model by fusing the CNN and RNN models, ensure that the dimensions of the feature vectors extracted by the CNN and RNN are consistent before connection.

[0067] Furthermore, a multi-season and multi-period prediction model is constructed by integrating CNN and RNN models, including:

[0068] S11. Connect the image feature vector extracted by CNN and the time series feature vector output by RNN to form a feature vector; wherein the connected feature vector will contain comprehensive information from the image and time series data.

[0069] S12. Establish a fully connected layer to process the fused feature vector through nonlinear transformation, and input the processed feature vector into the multi-season and multi-period prediction model; the fully connected layer can effectively process the fused feature vector and extract higher-level feature representation to provide support for the final prediction task. In the process of ship power consumption prediction, the fully connected layer helps the multi-season and multi-period prediction model learn to extract information useful for predicting power consumption from image and time series data.

[0070] S13. The predicted value of the ship's power consumption is output by calculating the forward propagation.

[0071] Furthermore, the prediction of the ship power consumption is calculated by the forward propagation output through the first neural network. Specifically, the fused feature vector is input into the fully connected layer, and the output of each fully connected layer is calculated through forward propagation, wherein the forward propagation calculation is expressed as: ;

[0072] in, It is The output vector of the layer, for layer 1 ; It is The weight matrix of the layer is expressed as ; It is The bias vector of the layer, dimension is , It is The activation function of the layer, is the predicted value of the neural network.

[0073] During the forward propagation process, the output of each layer depends on the output of the previous layer until the final output layer generates a predicted value.

[0074] According to the characteristics of ship electricity consumption, the multi-season and multi-period prediction model combined with CNN and RNN can comprehensively consider the multi-dimensional characteristics and time series characteristics of ship power consumption, provide more accurate and reliable power consumption prediction, and thus optimize the energy efficiency of ships.

[0075] S2. Extract spatiotemporal data through Monte Carlo method based on the predicted value, simulate the behavior of ships at different times and geographical locations, calculate the probability density function of ships at different travel and stay times, and then use Newton method to divide the charging probability of different time periods into multiple time scales to obtain the charging probability of ships at different time scales;

[0076] Furthermore, the spatiotemporal data are extracted through the Monte Carlo method based on the predicted values, and the behavior of ships at different times and geographical locations is simulated to calculate the probability density function of ships at different travel and stay times, including:

[0077] Set the value range of the time variable T and define the position of the ship and power consumption as a function of time and treated as a random variable;

[0078] According to T, and The probability distribution of the random sampling is used to generate simulation samples. Each simulation sample represents a ship behavior path. The more samples are taken, the more accurate the simulation results are.

[0079] By analyzing the simulated samples, the probability distribution of ships at different times and locations is obtained, and the path probability is calculated. as the frequency of occurrence of the path;

[0080] For continuous data, use the limit definition to calculate the vessel dwell time and ship travel time The probability density function of the specific residence time is expressed as:

[0081] ;

[0082] The probability density function of travel time is expressed as:

[0083] ;

[0084] The value range of the time variable T is set, and the position and power consumption of the ship are defined as functions of time and as random variables. Then, a large number of random sampling is performed based on the probability distribution of these variables to generate simulated samples representing various possible ship behavior paths. These samples are then analyzed to obtain the probability distribution of the ship at different times and positions, and the path probability, that is, the frequency of occurrence of a specific path, is calculated based on this. For continuous spatiotemporal data, the limit concept is used to calculate the probability density function of the ship's stay time and travel time, thereby obtaining the probability distribution of these two times. The results of this process can provide accurate probability information for ship operations, help optimize the ship's charging strategy, path planning and operation management, and ultimately achieve the goal of improving operational efficiency and reducing risks.

[0085] Furthermore, the Newton method is used to divide the charging probability of different time periods into multiple time scales, and the charging probability of the ship at different time scales is obtained. The specific contents include:

[0086] Set the first objective function , represents the expected value of the total power consumption of the ship at different time scales, and is a function of the charging probability P, where the first objective function is expressed as:

[0087] ;

[0088] in, It is the time period The weight of ) is in the time period The expected value of power consumption under It is in the time period 2 corresponding charging probability;

[0089] Use Newton's method to find the optimal charging probability so that the first objective function The optimal, specific Newton method iteration is expressed as: ,in is the charging probability at the kth iteration, The first objective function is The first-order derivative at , is the second-order derivative.

[0090] The charging probability of the ship at different time scales is divided by setting the first objective function. This first objective function comprehensively considers the expected value of power consumption in each time period and is expressed in the form of a weighted sum, where the weight reflects the importance of different time periods, and the charging probability directly affects the power consumption. Through the Newton method iteration process, a set of optimal charging probabilities can be effectively found. This set of probabilities can make the first objective function, that is, the sum of the expected values ​​of power consumption in different time periods, achieve the desired minimization or maximization effect. The efficient iterative characteristics of the Newton method can quickly converge to the optimal solution, thereby providing accurate decision-making support for the charging scheduling of ships in different time periods. The final result can not only optimize the power utilization efficiency of the ship and reduce energy consumption costs, but also improve the reliability and economy of ship operations.

[0091] Furthermore, according to the charging probability at different time scales, the spatial distribution and charging status of the ship during operation are clarified, and the charging load data are classified using fuzzy mathematics theorems and LightGBM, including: after clarifying the spatial distribution and charging status of the ship during operation, fuzzy mathematics theorems and LightGBM are used to classify and analyze the charging load data. Fuzzy mathematics processing fuzzifies the charging status of the ship in different time periods to generate fuzzy sets to cope with the uncertainty in the operation process. LightGBM classification classifies the charging load based on the ship's operating data and fuzzy processing results to predict the charging demand at different time scales.

[0092] S3. A multi-objective model is established based on the charging probability. The first objective function is solved by improving the non-dominated sorting genetic algorithm to obtain the power consumption strategy in different time periods. The multi-objective model is a multi-objective model with the goal of minimizing power consumption. It is to optimize the power consumption of ships at different time scales. The first objective function of the model is solved by improving the non-dominated sorting genetic algorithm (NSGA-II).

[0093] Furthermore, the first objective function is solved by improving the non-dominated sorting genetic algorithm to obtain the power consumption strategies for different time periods, including:

[0094] S31, randomly generate an initial population, each individual represents a potential electricity usage strategy, including the probability of charging in different time periods;

[0095] S32, performing non-dominated sorting on the current population, distinguishing individuals at each non-dominated level, and completing this by calculating the dominance number and dominated set of each individual;

[0096] S33. Maintain population diversity by calculating the crowding distance of each individual, select individuals based on their non-dominated rank and crowding degree, and give priority to individuals with lower ranks and larger crowding distances;

[0097] S34, combine selection, crossover and mutation operations to generate a new offspring population, merge the parent population with the offspring population, perform non-dominated sorting and crowding calculation again, and select a new parent population according to the elite strategy;

[0098] S35. Finally, a set of Pareto optimal solutions are generated, representing the optimal electricity consumption strategies in different time periods, reflecting the trade-offs among multiple objectives.

[0099] The NSGA-II algorithm can effectively solve multi-objective optimization problems, obtain the optimal power utilization strategy in different time periods, and optimize the ship's power utilization while considering multiple objectives such as power consumption and charging efficiency.

[0100] Specific power usage strategies include: short-term strategy: achieving instant charging at the hourly level to meet temporary high power demand, medium-term strategy: balancing charging and power consumption at the daily or weekly level to optimize the overall power management of the ship, and long-term strategy: planning power usage at the monthly level to reduce power consumption costs.

[0101] S4. Use the deep deterministic policy gradient (DDPG) algorithm to control ship power according to the power consumption strategy.

[0102] Furthermore, the deep deterministic policy gradient (DDPG) algorithm is used to control ship power according to the power consumption strategy, including:

[0103] S41, combining the predicted value with the current state information of the ship and inputting it into the DDPG algorithm as the current state;

[0104] S42, generating control instructions through the power strategy network, such as adjusting the power of equipment, changing load distribution, etc.;

[0105] Specific control instructions include: dynamically adjusting the power output of the generator set: adjusting the output power of the generator in real time according to the power load demand and forecast results to ensure the stability of the power supply; adjusting the power of the propulsion system: during navigation, adjusting the power output of the propulsion system according to factors such as the ship's speed and navigation conditions to achieve energy-saving effects; optimizing the power settings of auxiliary systems such as air conditioning and refrigeration: dynamically adjusting the power settings of these systems according to various meteorological environmental conditions (such as temperature, humidity) and the needs of the ship.

[0106] Specifically, based on the predicted power demand, reasonable scheduling of power generation and power distribution to avoid oversupply or shortage includes: Real-time power generation scheduling: According to the power demand forecast, the start and stop timing of the generator set is reasonably arranged to avoid losses caused by frequent start and stop; Battery management: For ships equipped with battery energy storage systems, instructions include storing electricity at low load and releasing electricity at high load to balance the power supply; Preventive maintenance plan: Based on the predicted equipment usage, equipment maintenance and inspection are arranged in advance to avoid power supply interruptions caused by sudden failures.

[0107] S43, evaluating the effectiveness of the generated control instructions through the value network, and optimizing the control decision by adjusting the parameters of the strategy network through SAC;

[0108] Specifically, the Bellman equation is used to calculate the action value to evaluate the effectiveness of the current control instruction. The action value is expressed as: ;

[0109] in, Indicates taking action at time t The expected return is parameterized by the value network as , Indicates the current state and actions The immediate reward (or reward function) under is the feedback obtained immediately after the agent performs an action. represents a discount factor that measures the value of future rewards relative to immediate rewards. If Close to 1, the agent will pay more attention to long-term rewards; if Close to 0, the agent will be more inclined to obtain immediate rewards; Indicates that at the next time point t+1 Take Action ; according to the calculated action value, when the calculated action value is higher than the expected return of the average level of the environment, it means that the control instruction of the action in the current state is valid; otherwise, it means that the action does not have a good return effect, and the control instruction of the action in the current state is invalid.

[0110] The action value means that the expected return of the current state and action is equal to the current immediate reward plus the present value of the expected future return (adjusted by the discount factor), which reflects the recursive nature of reinforcement learning, that is, the value of the current state depends not only on the current immediate reward, but also on the expected return that can be obtained by continuing to follow the strategy from the next state.

[0111] Furthermore, the parameters of the policy network are adjusted through SAC, which is specifically expressed as:

[0112] ;

[0113] in, Represents the gradient symbol, for the policy network parameters The gradient operator is used to find the second objective function The rising direction under the current parameters; represents the expected inner product, which is for the state The expectation of the state is sampled from the experience buffer D, which reflects the various states encountered by the policy network in its past interactions with the environment. represents the entropy regularization term, is a hyperparameter used to control the weight of the entropy term and balance exploration and utilization. is in state Under the policy network Select Action The logarithmic probability of .

[0114] The gradient expression is used to update the policy network parameters. By calculating the gradient of the second objective function relative to the parameters, the gradient points to the direction of the fastest growth of the second objective function. Therefore, in the SAC algorithm, the policy network parameters are updated according to the direction of growth to maximize the second objective function, that is, to improve the performance of the policy and increase exploration. In practice, this gradient will update the parameters of the policy network through the gradient ascent algorithm, thereby gradually optimizing the policy network and enabling it to obtain higher cumulative returns in the long run.

[0115] It should be explained that combining DDPG with SAC is an effective way to enhance the stability and exploration ability of the strategy. The introduction of SAC can make up for the lack of DDPG in exploration ability, making the model more adaptable and flexible in the complex ship power control environment. In a dynamically changing and highly uncertain environment (such as ship power control), the entropy regularization of SAC can avoid the strategy from being overly dependent on the current optimal solution, thereby improving the overall performance of the system. The ship power control system can better cope with complex and changing meteorological environmental conditions while maintaining efficient power management and energy saving. This combination has the potential to improve the robustness and stability of the control strategy in practical applications.

[0116] S44, sending the control instructions generated by the DDPG algorithm to the ship's power management system to perform specific power control operations.

[0117] It should be explained that the deep deterministic policy gradient (DDPG) algorithm is used for ship power control. The DDPG algorithm consists of a policy network (Actor) and a value network (Critic). The policy network generates control instructions (i.e. actions), while the value network evaluates the quality of the instructions (i.e. value), outputs the actions (control instructions) that should be taken in the current state, such as adjusting the power of the equipment or changing the operating mode of the ship system, and evaluates the quality of the given state and action pair, that is, the value of executing the action in the current state. By continuously adjusting the parameters of the policy network, the actions selected in each state can maximize the overall benefits of the ship's power system.

[0118] Specifically, by using the DDPG algorithm, not only can accurate control instructions be generated based on the predicted values, but the control strategy can also be continuously optimized through reinforcement learning to ensure the efficient and stable operation of the ship's power system. This process is suitable for complex ship power control tasks and can handle the challenges of multi-dimensional state space and continuous action space, thereby realizing intelligent control of ship power consumption.

[0119] It should be noted that the first objective function focuses on minimizing the power consumption in different time periods, expressing the comprehensive consumption through weighted sums, with the emphasis on the division of time scales and the determination of the optimal charging probability. The second objective function is related to DDPG, with the goal of maximizing long-term returns and optimizing the effectiveness and efficiency of specific operations, such as the control and scheduling of ship power systems.

[0120] In this embodiment, a multi-season and multi-period prediction model is first established. The model uses a hybrid structure of a convolutional neural network (CNN) and a recurrent neural network (RNN) to process ship operating data, effectively capturing the time series characteristics and complex patterns of power consumption. Then, the Monte Carlo method is used to combine the spatiotemporal data of the ship for dynamic simulation, and the probability density function and Newton method are used to optimize the charging probability distribution under multiple time scales. In addition, the present invention also constructs a multi-objective optimization model and uses an improved non-dominated sorting genetic algorithm (NSGA-II) to solve it to generate power consumption strategies for different time periods. Finally, the deep deterministic policy gradient (DDPG) algorithm and the soft actor-critic (SAC) method are used to achieve precise regulation of ship power. This series of processes not only optimizes the energy efficiency of the ship, but also improves the reliability and economy of the operation, providing a new solution for intelligent power management of ships.

[0121] Example 2

[0122] This embodiment provides a ship intelligent power consumption system based on neural network, which is applied in a ship intelligent power consumption method based on neural network, including a data receiving and prediction module, a Monte Carlo simulation and Newton method optimization module, an intelligent power consumption strategy optimization module and a power regulation and control instruction execution module.

[0123] The data receiving and prediction module is used to receive ship operating condition data from a data source, establish a multi-season and multi-period prediction model, receive information as an input vector, and output a predicted value of ship power consumption using a neural network;

[0124] The Monte Carlo simulation and Newton method optimization module uses the Monte Carlo method to extract spatiotemporal data, simulate ship behavior, calculate the probability density function of ships under different travel and stay times, and use the Newton method to divide the multi-time scale charging probability in different time periods;

[0125] The intelligent power consumption strategy optimization module establishes a multi-objective model based on the charging probability, solves the second objective function by improving the non-dominated sorting genetic algorithm, generates a Pareto optimal solution, and obtains the optimal power consumption strategy for different time periods;

[0126] The power regulation and control instruction execution module combines the power consumption strategy and the deep deterministic policy gradient algorithm to perform power regulation, uses the SAC algorithm to adjust the policy network parameters, optimizes the control decision, and sends the generated control instructions to the ship's power management system for execution.

[0127] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for intelligent power consumption of ships based on neural network, characterized in that: The following steps are involved: receiving marine operating condition data from a data source, establishing a multi-season and multi-period prediction model, and transmitting the received information as an input vector to a neural network of the multi-season and multi-period prediction model; Outputting a predicted value of ship power consumption according to the neural network; According to the predicted value, the spatiotemporal data is extracted by Monte Carlo method, the behavior of the ship at different times and geographical locations is simulated, and the probability density function of the ship at different travel and stay times is calculated. Then, the Newton method is used to divide the charging probability of different time periods into multiple time scales, and the charging probability of the ship at different time scales is obtained; The method of extracting spatiotemporal data by Monte Carlo method according to the predicted value, simulating the behavior of the ship at different times and geographical locations, and calculating the probability density function of the ship at different travel and stay times includes: Set the value range of the time variable T and define the position of the ship and power consumption as a function of time and treated as a random variable; According to T, and Random sampling is performed on the probability distribution to generate simulated samples; By analyzing the simulated samples, the probability distribution of ships at different times and locations is obtained, and the path probability is calculated. as the frequency of occurrence of the path; For continuous data, the probability density functions of ship stay times and ship travel times are calculated using the limiting definition; The probability density function of the residence time is expressed as: ;in, represents the ship's stay time, and the probability density function of the travel time is expressed as: ; Indicates the ship's travel time; The Newton method is used to divide the charging probability of different time periods into multiple time scales, and the charging probability of the ship at different time scales is obtained, including: Set the first objective function , represents the expected value of the total power consumption of the ship at different time scales, and is a function of the charging probability P, where the first objective function is expressed as: ； in, It is the time period The weight of ) is in the time period The expected value of power consumption under It is in the time period The corresponding charging probability is: Use Newton's method to find the optimal charging probability so that the first objective function The optimal, specific Newton method iteration is expressed as: ,in is the charging probability at the kth iteration, The first objective function is The first-order derivative at , is the second-order derivative; It also includes using a deep deterministic policy gradient algorithm to control ship power according to the power consumption strategy; wherein the predicted value is combined with the current state information of the ship and input into the DDPG algorithm as the current state; Generate control instructions through the power strategy network; The effectiveness of the generated control instructions is evaluated through the value network, and the parameters of the strategy network are adjusted through SAC to optimize the control decision; The control instructions generated by the DDPG algorithm are sent to the ship's power management system to perform power regulation operations.

2. The method for intelligent electricity use of ships based on neural network according to claim 1, characterized in that: By integrating CNN and RNN models, a multi-season and multi-period forecasting model is constructed, including: Connect the image feature vector extracted by CNN and the time series feature vector output by RNN to fuse them into a feature vector; Establish a fully connected layer to process the fused feature vector through nonlinear transformation, and input the processed feature vector into the multi-season and multi-period prediction model; The predicted value of the ship's power consumption is output by calculating the forward propagation.

3. The method for intelligent electricity use of ships based on neural network according to claim 2, characterized in that: The prediction of ship power consumption is calculated by forward propagation output through the first neural network. Specifically, the fused feature vector is input into the fully connected layer, and the output of each fully connected layer is calculated through forward propagation, where the forward propagation calculation is expressed as: ； in, It is The output vector of the layer, for layer 1 ; It is The weight matrix of the layer is expressed as ; It is The bias vector of the layer, dimension is , It is The activation function of the layer, is the predicted value of the neural network; During the forward propagation process, the output of each layer is based on the output of the previous layer until the final output layer generates a predicted value.

4. The method for intelligent power consumption of ships based on neural network according to claim 1, characterized in that: The method also includes establishing a multi-objective model according to the charging probability, solving the first objective function by improving the non-dominated sorting genetic algorithm, and obtaining the electricity utilization strategy in different time periods.

5. The method for intelligent electricity use of ships based on neural network according to claim 4 is characterized in that: The first objective function is solved by improving the non-dominated sorting genetic algorithm to obtain the electricity consumption strategies in different time periods, including: The initial population is randomly generated, and each individual represents a potential electricity usage strategy, including the probability of charging in different time periods; Perform non-dominated sorting on the current population, and distinguish individuals at each non-dominated level by calculating the dominance number and dominated set of each individual; The population diversity is maintained by calculating the crowding distance of each individual, and individuals are selected based on their non-dominated rank and crowding degree; Combine selection, crossover and mutation operations to generate a new offspring population, merge the parent population with the offspring population, perform non-dominated sorting and crowding calculation again, and select a new parent population based on the elite strategy; Generate a set of Pareto optimal solutions, representing the optimal electricity consumption strategies in different time periods.

6. The method for intelligent power consumption of ships based on neural network according to claim 1, characterized in that: The Bellman equation is used to calculate the action value to evaluate the effectiveness of the current control instruction. The action value is expressed as: ; in, Indicates taking action at time t The expected return is parameterized by the value network as , Indicates the current state and actions The immediate reward is the feedback obtained immediately after the agent performs an action. represents the discount factor, which is used to measure the value of future rewards relative to immediate rewards, Indicates that at the next time point t+1 Take Action expected return; when the calculated action value is higher than the average level of the environment, it means that the control instruction of the action in the current state is valid; otherwise, the control instruction of the action in the current state is invalid.

7. The method for intelligent power consumption of ships based on neural network according to claim 6 is characterized in that: The parameters of the policy network are adjusted through SAC, which is specifically expressed as: ； in, Represents the gradient symbol, for the policy network parameters The gradient operator is used to find the second objective function The rising direction under the current parameters; represents the expected inner product, which is for the state The expectation of the state is sampled from the experience buffer D. represents the entropy regularization term, is a hyperparameter that controls the weight of the entropy term, is in state Under the policy network Select Action The logarithmic probability of .

8. A ship intelligent power consumption system based on a neural network, applied to a ship intelligent power consumption method based on a neural network as claimed in any one of claims 1 to 7, characterized in that: It includes data receiving and prediction module, Monte Carlo simulation and Newton method optimization module, intelligent power consumption strategy optimization module and power regulation and control instruction execution module. The data receiving and prediction module is used to receive ship operating condition data from a data source, establish a multi-season and multi-period prediction model, receive information as an input vector, and output a predicted value of ship power consumption using a neural network; The Monte Carlo simulation and Newton method optimization module uses the Monte Carlo method to extract spatiotemporal data, simulate ship behavior, calculate the probability density function of ships under different travel and stay times, and use the Newton method to divide the multi-time scale charging probability in different time periods; The intelligent power consumption strategy optimization module establishes a multi-objective model based on the charging probability, solves the second objective function by improving the non-dominated sorting genetic algorithm, generates a Pareto optimal solution, and obtains the optimal power consumption strategy for different time periods; The power regulation and control instruction execution module combines the power consumption strategy and the deep deterministic policy gradient algorithm to perform power regulation, uses the SAC algorithm to adjust the policy network parameters, optimizes the control decision, and sends the generated control instructions to the ship's power management system for execution.

Images