A short-term load forecasting method and system for a ship's electric propulsion system

Abstract

The present invention belongs to the technical field of electric load forecasting, and specifically relates to a short-term load forecasting method and system for a ship electric propulsion system, comprising: acquiring operating data of the ship electric propulsion system; performing data preprocessing on the acquired operating data; extracting time series features of the preprocessed ship electric propulsion system operating data through a time series convolutional network to obtain a time series feature sequence of the ship electric propulsion system operation; and performing short-term load forecasting of the ship electric propulsion system based on the obtained time series feature sequence and a bidirectional long short-term memory network.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric load forecasting, and in particular relates to a short-term load forecasting method and system for a ship electric propulsion system. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the development of society, hydrogen ships are gradually receiving more attention. Figure 1 As shown in the figure, compared with ordinary diesel ships, hydrogen ships have lower power generation capacity, lower system inertia, and weaker self-regulation capabilities. Furthermore, the power electronics and high-power energy storage devices make hydrogen ships more demanding on power quality. In contrast, the propulsion load of hydrogen ships generally accounts for over 70% of the total ship capacity. During navigation, wind and waves, as well as the ship's starting, stopping, and turning, can cause fluctuations in propulsion load, resulting in frequency and voltage oscillations, reducing power quality and causing energy waste. Therefore, predicting propulsion load changes can provide a reference for controlling the ship's electric propulsion system, thereby protecting the ship's electrical and mechanical systems and extending its lifespan.

[0004] Load forecasting can be categorized into short-term, medium-term, and long-term load forecasting based on timeframe. Minute-level short-term load forecasting (STLF) enables real-time prediction of propulsion load changes during ship navigation, allowing for flexible adjustment of system control strategies and reducing energy loss. Currently, short-term load forecasting is widely used in land power systems. Prediction methods primarily include traditional statistical models and machine learning models. Statistical models include exponential smoothing (ETS), autoregressive integrated moving average (ARIMA), gray models (GM), and linear regression (LR). These methods can meet the requirements of linear forecasting, but they fail to account for the nonlinear characteristics of load series, resulting in low forecast accuracy for long-term series. To accurately handle nonlinear loads, machine learning models are widely used. These include gradient regression tree (GBRT), random forest (RF), and least squares support vector machine (LSSVM). In recent years, neural network-based deep learning, due to its advantages in processing large-scale, nonlinear sequences, has been increasingly applied to load forecasting. Deep learning models include long short-term memory (LSTM), gated recurrent unit (GRU), and deep belief network (DCN). However, a single deep learning model currently has the disadvantages of limited applicability and low prediction accuracy.

[0005] According to the inventors, hybrid prediction models have been applied to load forecasting. Although there are many advanced load prediction algorithms, the high proportion of propulsion load, high dynamics, randomness, and high environmental dependence of ship power systems make load prediction less widely used. In addition, ships are subject to various operating conditions during navigation, such as operating conditions, port entry and exit conditions, and emergency conditions. The propulsion load varies greatly between different operating conditions, which will seriously affect the prediction accuracy of the propulsion load. Currently, there are few studies on propulsion load prediction for ship power systems, and some studies are not complete. Summary of the Invention

[0006] In order to solve the above problems, the present invention proposes a short-term load forecasting method and system for a ship electric propulsion system, which adopts a neural network model that combines a temporal convolutional neural network (TCN) and a bidirectional long short-term memory neural network (BiLSTM) to predict the short-term power load. The short-term load forecasting of the ship electric propulsion system is performed based on multi-step feature processing, thereby improving the prediction performance and the prediction accuracy.

[0007] According to some embodiments, a first solution of the present invention provides a short-term load forecasting method for a ship electric propulsion system, which adopts the following technical solution:

[0008] A short-term load forecasting method for a ship electric propulsion system, comprising:

[0009] Obtain operational data of ship electric propulsion systems;

[0010] Performing data preprocessing on the acquired operating data;

[0011] The time series features of the pre-processed ship electric propulsion system operation data are extracted through the time series convolutional network to obtain the time series feature sequence of the ship electric propulsion system operation;

[0012] The short-term load forecasting of the ship electric propulsion system is carried out based on the obtained time series feature sequence and bidirectional long short-term memory network.

[0013] As a further technical limitation, the obtained time series feature sequence is input into a bidirectional long short-term memory network, and the forward and reverse information of the input time series feature sequence is learned at the same time to explore the potential time correlation of the load of the ship's electric propulsion system, and complete the short-term load prediction under the action of the fully connected layer of the bidirectional long short-term memory network.

[0014] As a further technical limitation, the hidden information and data features in the pre-processed ship electric propulsion system operation data are mined through a time series convolutional network, and the mined hidden information and data features are reconstructed to obtain a time series feature sequence.

[0015] As a further technical limitation, the data preprocessing at least includes variational mode decomposition, principal component analysis and K-means clustering.

[0016] Furthermore, the operation data of the ship electric propulsion system obtained by variational mode decomposition is processed to obtain the eigenmode group containing the component characteristics of the operation data. The data dimensionality reduction is performed through principal component analysis to retain the main components of the data; K-means mean clustering is used to automatically classify the data to obtain the preprocessed ship electric propulsion system operation data.

[0017] As a further technical limitation, the temporal convolutional network adopts a residual connection structure, which forms a residual module by connecting deep neural networks across layers, and sets several residual modules after the convolutional layer to avoid gradient vanishing and gradient explosion caused by neural network degradation.

[0018] According to some embodiments, a second solution of the present invention provides a short-term load forecasting system for a ship electric propulsion system, which adopts the following technical solution:

[0019] A short-term load forecasting system for a ship electric propulsion system, comprising:

[0020] an acquisition module configured to acquire operating data of a ship electric propulsion system;

[0021] a preprocessing module, configured to perform data preprocessing on the acquired operation data;

[0022] an extraction module configured to extract the time series features of the preprocessed ship electric propulsion system operation data through a time series convolutional network to obtain a time series feature sequence of the ship electric propulsion system operation;

[0023] The prediction module is configured to perform short-term load prediction of the ship electric propulsion system based on the obtained time series feature sequence and the bidirectional long short-term memory network.

[0024] According to some embodiments, a third solution of the present invention provides a computer-readable storage medium, which adopts the following technical solution:

[0025] A computer-readable storage medium stores a program thereon, which, when executed by a processor, implements the steps of the short-term load forecasting method for a ship electric propulsion system as described in the first embodiment of the present invention.

[0026] According to some embodiments, a fourth solution of the present invention provides an electronic device, which adopts the following technical solution:

[0027] An electronic device comprises a memory, a processor and a program stored in the memory and running on the processor, wherein when the processor executes the program, the steps of the short-term load forecasting method for a ship electric propulsion system as described in the first embodiment of the present invention are implemented.

[0028] According to some embodiments, a fifth solution of the present invention provides a computer program product, which adopts the following technical solution:

[0029] A computer program product includes software codes, wherein the program in the software codes executes the steps of the short-term load forecasting method for a ship electric propulsion system according to the first embodiment of the present invention.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] The present invention uses a TCN-BiLSTM hybrid prediction method to perform short-term load forecasting of a ship's electric propulsion system. While ensuring the training gradient, it effectively extracts the deep time series features and hidden information of the ship during operation. It uses BiLSTM to process time series features, mines the forward and backward time series associations in the feature sequence, and performs short-term load forecasting of the ship's electric propulsion system based on multi-step feature processing, thereby improving both prediction performance and prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The drawings constituting a part of the specification of this embodiment are used to provide a further understanding of this embodiment. The schematic embodiments and descriptions of this embodiment are used to explain this embodiment and do not constitute an improper limitation on this embodiment.

[0033] Figure 1 Schematic diagram of the structure of a ship electric propulsion system in the background technology of the present invention;

[0034] Figure 2 This is an architectural diagram of a short-term load forecasting method for a ship electric propulsion system according to a first embodiment of the present invention;

[0035] Figure 3 This is a flow chart of data preprocessing in Example 1 of the present invention;

[0036] Figure 4 Schematic diagram of the convolution structure in the temporal convolutional network in Example 1 of the present invention;

[0037] Figure 5 Schematic diagram of the structure of residual connection in the temporal convolutional network in embodiment 1 of the present invention;

[0038] Figure 6 Schematic diagram of the structure of a bidirectional long short-term memory network in embodiment 1 of the present invention;

[0039] Figure 7Schematic diagram of ship operation data in embodiment 1 of the present invention; wherein, Figure 7 (a) is a schematic diagram of ship propulsion power; Figure 7 (b) is a schematic diagram of the ship's motor speed; Figure 7 (c) is a schematic diagram of wind speed; Figure 7 (d) is a schematic diagram of the output power of the ship's battery; Figure 7 (e) is a schematic diagram of the ship's power battery; Figure 7 (f) is a schematic diagram of daily load data;

[0040] Figure 8 Schematic diagram of ship load prediction results in Example 1 of the present invention; wherein, Figure 8 (a) is a load prediction diagram of a ship in normal operating conditions when the sampling point unit is 1 minute; Figure 8 (b) is a load prediction diagram of the ship in operating condition when the sampling point unit is 1 minute; Figure 8 (c) is a load prediction diagram of a ship entering or leaving a port when the sampling point unit is 1 minute; Figure 8 (d) is a load prediction diagram of the ship in normal operating conditions when the sampling point unit is 2 minutes; Figure 8 (e) is a load prediction diagram of the ship in the operating condition when the sampling point unit is 2 minutes; Figure 8 (f) is a load prediction diagram of a ship entering or leaving a port when the sampling unit is 2 minutes; Figure 8 (h) is a load prediction diagram of the ship in normal operating conditions when the sampling point unit is 5 minutes; Figure 8 (i) is a load prediction diagram of the ship in the operating condition when the sampling point unit is 5 minutes; Figure 8 (j) in the figure is a load prediction diagram of a ship entering or leaving a port when the sampling point unit is 5 minutes. DETAILED DESCRIPTION

[0041] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0042] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present application belongs.

[0043] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0044] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0045] Example 1

[0046] Embodiment 1 of the present invention introduces a short-term load forecasting method for a ship electric propulsion system.

[0047] like Figure 2 As shown, this embodiment performs data preprocessing on the acquired operating data, extracts the time series feature sequence based on the time series convolutional network, and completes the short-term load forecast of the ship's electric propulsion system through a bidirectional long short-term memory network.

[0048] This embodiment adopts Figure 3 The method shown is used to preprocess the acquired ship electric propulsion system operation data. In order to solve the high discreteness and high noise characteristics of the data, variational mode decomposition (VMD) is used to convert the mixed data into a smooth and periodic intrinsic mode group. The use of VMD will greatly increase the dimension and scale of the data, which will increase the complexity of the algorithm and is not conducive to the timeliness of data processing. Therefore, the PCA method is used to extract the main features of the data to the greatest extent, and reduce the data dimension and scale while ensuring the least information loss. In order to solve the problem of lack of correlation between the internal features of the processed information, the K-means clustering method is used to divide the data into different categories according to the internal feature connection of the data, thereby enhancing the readability of the data. Data processing reduces the volatility and randomness of the data, and can adapt to complex ship propulsion operation conditions to improve the model prediction performance.

[0049] The variational mode decomposition used in this embodiment is an adaptive, completely non-recursive signal processing algorithm that can decompose complex signals into multiple physically meaningful eigenmodes and has solid mathematical theoretical support. Compared with traditional signal decomposition methods such as empirical mode decomposition, the VMD in this embodiment can effectively solve the problem of modal aliasing of the decomposed signal by controlling the bandwidth, which is more conducive to processing highly redundant ship operation data.

[0050] The variational mode decomposition process in this embodiment is as follows:

[0051] (1) Construct an equivalent unconstrained variational problem, introduce the Lagrange multiplier operator λ and the quadratic penalty factor α, and the augmented Lagrange expression can be obtained as follows: where K is a positive integer representing the number of modes to be decomposed; {u k} = {u1,..., u K}, {ω k} = {ω1,…, ω K} represent the set of K decomposed mode components and their corresponding central frequencies; δ(t) is the Dirac function; * is the convolution operator; f is the input signal.

[0052] (2) Use the alternating direction multiplier method, Fourier isometric transform to solve for each mode component and its central frequency, that is where n is the number of iterations, γ is the noise tolerance, is the form after Fourier transform. The variational mode decomposition in this embodiment will be used to decompose highly mixed data.

[0053] The principal component analysis PCA adopted in this embodiment is a widely used unsupervised data dimensionality reduction algorithm, which maps n-dimensional features to k dimensions (k < n) through linear transformation. The transformed k-dimensional data is the principal component of the original n-dimensional features. VMD decomposes a set of feature data into the superposition of multiple sets of modal data, realizes the information difference in the data, and increases the data dimension. PCA measures the difference of data by selecting the reference variance, maps the high-dimensional data with large differences to low dimensions, and extracts the main feature information of the data after VMD decomposition; assume the input n-dimensional data set X = {x1, x2,…, x n}, which needs to be reduced to k dimensions; specifically, the data set is de-centered to obtain the covariance matrix Use the method of eigenvalue decomposition to find the eigenvalues and eigenvectors of the covariance matrix C; sort the eigenvalues and eigenvectors, select the largest k, and recombine their corresponding eigenvectors into a row vector group P; transform the data into the new space constructed by k eigenvectors, that is Y = PX.

[0054] Unlike classification, clustering is a form of unsupervised learning that automatically groups similar samples into one category based on the characteristics within the data without sample labels. K-means clustering is a commonly used clustering algorithm. After determining the K value and the K initial cluster centers, it traverses all data points and divides them into the nearest cluster. After the traversal is completed, the distance from all data points in a cluster to the cluster center is averaged, and the cluster center is iteratively updated until a certain termination condition is reached. The K-means method can divide the processed data into different categories and label each category with a "label", which makes the data features more obvious and more readable, improves the accuracy of model predictions, and reduces the time for prediction learning.

[0055] The steps of k-means clustering used in this embodiment are:

[0056] Assume that the sample data X={x1,x2,…,x n} contains n objects, each of which has m dimensional attributes; initialize k cluster centers Define the loss function as Let t = 0, 1, 2...n be the number of iterations, and repeat the following process until the loss function converges: For each sample x i , assign it to the nearest center For each cluster, recalculate the center point of the cluster Clustering is completed.

[0057] This embodiment adopts Figure 4 and Figure 5 The temporal convolutional network TCN (Temporal Convolutional Network) shown is improved from the convolutional neural network (CNN). The TCN module will perform feature extraction on the processed time series data. The TCN network in this embodiment contains multiple residual block structures, which effectively avoids the problems of gradient vanishing and gradient explosion that may occur when processing large-scale data. Each residual module contains dilated convolution and causal convolution, Dropout mechanism, activation function and weight normalization mechanism. Dropout is a regularization technology that helps the prediction model improve its generalization ability and solves the overfitting problem by randomly ignoring a part of neurons during the training process. As an important feature of the neural network, the activation function introduces nonlinear features to map the input of the neural network to the output.

[0058] Causal convolution and dilated convolution, residual structure is the key feature of TCN, specifically:

[0059] 1) Causal convolution: The value of the previous layer at time t only depends on the value of the next layer at time t and before. That is, causal convolution cannot see future data. It is a one-way structure and a strict time constraint model. t The output result at time t depends only on the input x at the current time. t and part of the past time input (x t-1 ,x t-2 ,x t-3 …), and the excess part is padded with zeros to ensure that the input and output sequences have the same length.

[0060] 2) Dilated convolution: To solve the problem that causal convolution requires many layers or large convolution kernels / filters to increase the receptive field, TCN introduces dilated convolution. Dilated convolution can skip part of the input so that the filter can be applied to an area larger than its own length, thereby multiplying the receptive field of the network. For a one-dimensional sequence input x∈R n And the filter f:{0,…,k-1}→R, the dilated convolution is Among them, k is the size of the convolution kernel, d is the order of 2

[0061] Activation(·)

[0062] Multiplication is the expansion factor, and the receptive field of each layer is (k-1)d.

[0063] 3) Residual structure: To avoid gradient vanishing and gradient exploding caused by neural network degradation, residual connections are used in TCN, which form residual blocks by connecting deep neural networks across layers. Figure 5 The residual block shown in the figure introduces the activation function ReLu on top of the convolutional layer to ensure the nonlinearity of the model. The activation function is added to the residual block after the convolutional layer to normalize the input of the hidden layer and apply weight normalization to each convolutional layer. To prevent overfitting, regularization is introduced through dropout. The extraction of temporal features is completed based on o = Activation(x + F(x)). Where o represents the output, F() is the dilated convolution, and Activation(·) is the activation function.

[0064] The bidirectional long short-term memory neural network (BiLSTM) in this embodiment is a variant of LSTM (Long Short-Term Memory), consisting of a forward LSTM layer and a backward LSTM layer. Its basic structural unit is LSTM. An LSTM unit mainly includes a forget gate, a memory gate, and an output gate; t ,h t represents the input and output of the hidden layer at time t, represents the state candidate vector in the memory process; f tThe forget gate receives input and the output of the previous unit through the sigmoid layer to decide which information to discard; t Represents the memory gate, using a sigmod layer and a tanh layer, through i t and Multiply and select the newly added information and update the LSTM unit state; the output gate is o t , by x t , h t-1 Output through the sigmoid layer and C after the tanh function t , and get the output signal h t .

[0065] like Figure 6 As shown in the figure, BiLSTM connects the input sequence to a forward LSTM network and a reverse LSTM network respectively, merges the outputs of the two LSTMs, and thus obtains bidirectional information in the input sequence at the same time. After TCN extraction, the ship time series features contain a large amount of information. The bidirectional extraction capability of BiLSTM can predict the time series association between the past and the future. Mining the hidden information helps to improve the prediction accuracy.

[0066] This example connects the feature vectors output by the BiLSTM to a 128-node fully connected layer using the Relu activation function to obtain the final load forecast result. Specifically, an operational dataset for electric-powered ships is imported, containing data such as propulsion load power, daily load, power of power batteries and storage batteries, and ship speed. The dataset contains 42,000 data points with a sampling interval of 1 minute. The dataset is divided into training, test, and validation sets in an 8:1:1 ratio and normalized. Data preprocessing parameters are set, including the number of modes, center frequency, and noise tolerance in VMD, the number of principal components in PCA, and the number of clusters in k-means. Hybrid prediction model parameters are set, including the number of layers, filter size and number, and expansion coefficient of the TCN network, the number of layers and nodes in the BiLSTM network, and the number of layers and nodes in the fully connected layer. A data processing framework is created, using the data from the training set as input.

[0067] like Figure 3 As shown in the figure, the three algorithms are connected in series and a multi-step fusion method is used to transform the data into multiple meaningful clusters as the input of the prediction model; Figure 2The TCN-BiLSTM hybrid prediction model shown in the figure includes an input layer, a TCN layer, a BiLSTM layer, a fully connected layer, and an output layer structure. Based on the processed data, the time series features of the ship operation data are extracted through a TCN layer containing 128 second-order filters, and two-way feature prediction is achieved through a 4-layer BiLSTM layer. The fully connected layer composed of a 128-node ReLU activation function outputs the prediction result. The network training parameters are set, the learning rate is set to 0.05-0.1, the time window length is 20-50, and the epoch is 50-100. The overall prediction framework is trained and training is stopped when the loss function stabilizes. The above operations are repeated and the prediction step size is changed to complete the load forecast for the next 1 minute, 2 minutes, and 5 minutes.

[0068] It should be noted that if Figure 7 As shown, the ship driving data set in this embodiment contains a variety of ship operating conditions. There are large differences between each operating condition. According to different characteristics, it is divided into normal ship driving conditions, ship operation conditions, and ship entering and leaving the port conditions for error evaluation. The load prediction results are shown in Figure 2. Figure 8 shown.

[0069] This embodiment selects mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) to evaluate the prediction effect of the model, and obtains the error data table shown in Table 1; specifically,

[0070] Where n represents the number of samples in the dataset, and y i They represent the i-th predicted value and actual data respectively; MAE can intuitively reflect the prediction accuracy of the model; MAPE reflects the relative error between the data set and the predicted data in the form of a percentage; RMSE is the square root of the root mean square error, which can reflect the standard deviation of the residual.

[0071] Table 1 Error data table

[0072]

[0073] According to Table 1, Figure 7 and Figure 8, it can be seen that the prediction framework in this embodiment can achieve the best prediction effect under different operating conditions and different prediction times. Based on the data in Table 1, it can be seen that as the time scale increases, the prediction framework error in this embodiment also increases slightly; under normal driving conditions, the relatively stable propulsion load changes achieve the best prediction effect; under operating conditions, the ship maintains a high propulsion power output and is accompanied by long-term fluctuations. Compared with normal conditions, the prediction error under different time scales increases to varying degrees; compared with operating conditions, the propulsion load of the ship under port entry and exit conditions has a greater frequency of change, and the peaks and troughs are more obvious. The method in this embodiment has high robustness and ensures prediction accuracy under drastically fluctuating conditions. At the same time, due to the general law of load prediction, when the prediction step increases, the prediction error will also increase accordingly.

[0074] Example 2

[0075] The second embodiment of the present invention introduces a short-term load forecasting system for a ship electric propulsion system.

[0076] A short-term load forecasting system for a ship electric propulsion system, comprising:

[0077] an acquisition module configured to acquire operating data of a ship electric propulsion system;

[0078] a preprocessing module, configured to perform data preprocessing on the acquired operation data;

[0079] an extraction module configured to extract the time series features of the preprocessed ship electric propulsion system operation data through a time series convolutional network to obtain a time series feature sequence of the ship electric propulsion system operation;

[0080] The prediction module is configured to perform short-term load prediction of the ship electric propulsion system based on the obtained time series feature sequence and the bidirectional long short-term memory network.

[0081] The detailed steps are the same as those of the short-term load forecasting method for the ship electric propulsion system provided in Example 1 and will not be repeated here.

[0082] Example 3

[0083] A third embodiment of the present invention provides a computer-readable storage medium.

[0084] A computer-readable storage medium stores a program thereon, which, when executed by a processor, implements the steps of the short-term load forecasting method for a ship electric propulsion system as described in the first embodiment of the present invention.

[0085] The detailed steps are the same as those of the short-term load forecasting method for the ship electric propulsion system provided in Example 1 and will not be repeated here.

[0086] Example 4

[0087] A fourth embodiment of the present invention provides an electronic device.

[0088] An electronic device comprises a memory, a processor, and a program stored in the memory and running on the processor. When the processor executes the program, the steps of the short-term load forecasting method for a ship electric propulsion system as described in the first embodiment of the present invention are implemented.

[0089] The detailed steps are the same as those of the short-term load forecasting method for the ship electric propulsion system provided in Example 1 and will not be repeated here.

[0090] Example 5

[0091] A fifth embodiment of the present invention provides a computer program product.

[0092] A computer program product includes software code, wherein the program in the software code executes the steps of the short-term load forecasting method for a ship electric propulsion system as described in the first embodiment of the present invention.

[0093] The detailed steps are the same as those of the short-term load forecasting method for the ship electric propulsion system provided in Example 1 and will not be repeated here.

[0094] The above description is merely a preferred embodiment of this embodiment and is not intended to limit this embodiment. Those skilled in the art will readily appreciate that this embodiment may be modified and varied in various ways. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this embodiment shall be within the scope of protection of this embodiment.

Claims

1. A short-term load forecasting method for a ship electric propulsion system, characterized in that: include: Obtain operational data of ship electric propulsion systems; Performing data preprocessing on the acquired operating data; The temporal features of the pre-processed ship electric propulsion system operation data are extracted through a temporal convolutional network to obtain the temporal feature sequence of the ship electric propulsion system operation; Based on the obtained time series feature sequence and the bidirectional long short-term memory network, the short-term load of the ship's electric propulsion system is predicted. The obtained time series feature sequence is input into the bidirectional long short-term memory network, and the forward and reverse information of the input time series feature sequence is learned at the same time to explore the potential time correlation of the load of the ship's electric propulsion system. The short-term load prediction is completed under the action of the fully connected layer of the bidirectional long short-term memory network. The hidden information and data features in the pre-processed ship electric propulsion system operation data are mined through a time series convolutional network, and the mined hidden information and data features are reconstructed to obtain a time series feature sequence; The operation data of the ship electric propulsion system obtained by variational mode decomposition is processed to obtain the eigenmode group containing the component characteristics of the operation data. The data dimension reduction is processed by principal component analysis to retain the main components of the data. K-means clustering is used to automatically classify the data to obtain the preprocessed ship electric propulsion system operation data. The temporal convolutional network adopts a residual connection structure, which forms a residual module by connecting deep neural networks across layers. Several residual modules are set after the convolutional layer to avoid gradient vanishing and gradient explosion caused by neural network degradation.

2. A short-term load forecasting system for a ship electric propulsion system, which adopts the short-term load forecasting system method for a ship electric propulsion system according to claim 1, characterized in that: include: an acquisition module configured to acquire operating data of a ship electric propulsion system; a preprocessing module, configured to perform data preprocessing on the acquired operation data; an extraction module configured to extract the time series features of the preprocessed ship electric propulsion system operation data through a time series convolutional network to obtain a time series feature sequence of the ship electric propulsion system operation; The prediction module is configured to perform short-term load prediction of the ship electric propulsion system based on the obtained time series feature sequence and the bidirectional long short-term memory network.

3. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the short-term load forecasting method for a ship electric propulsion system as claimed in claim 1 are realized.

4. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the program, the steps of the short-term load forecasting method for a ship electric propulsion system as claimed in claim 1 are implemented.

5. A computer program product comprising software code, characterized in that The program in the software code executes the steps of the short-term load forecasting method for a ship electric propulsion system as claimed in claim 1.

Images