A power system probabilistic load forecasting method, system, device and storage medium

Abstract

This application discloses a method, system, device, and storage medium for probabilistic load prediction of power systems, relating to the field of load prediction technology for hydrogen-powered ship power systems. It addresses the technical problems in existing technologies, such as the lack of integration of multi-source data and physical information rules, and the failure to consider multi-scale chaos under the complex operating environment of HPV. Specifically, it includes: multi-source data acquisition and preprocessing; filtering strongly correlated data using the maximum information coefficient, and reconstructing multi-source data into a unified spatiotemporal feature matrix; constructing an electrical and environmental parametric chaotic model based on the Lorentz dynamic equation as a physical constraint term in the loss function; and building a BERT-PINN framework with an attention mechanism, achieving multi-step probabilistic prediction through two-stage training. This application effectively integrates the data-driven model and physical information rules of multi-source data through the above methods, and solves the feature fusion problem of multi-scale chaotic characteristics, thereby improving the model's predictive performance.

Description

Technical Field

[0001] This invention belongs to the field of load prediction technology for hydrogen-powered ship power systems, specifically relating to a method, system, equipment, and storage medium for probabilistic load prediction of power systems. Background Technology

[0002] Currently, hydrogen-powered ships (HPVs) are a key pathway to carbon neutrality in the shipping industry. Their Power System Load Forecasting (PSLF) is crucial for energy management, fuel cell lifetime optimization, and hydrogen supply scheduling. However, HPV power load exhibits multi-scale coupling characteristics, and the navigation environment and energy system exhibit time-varying chaotic correlations. Existing land-based microgrid PSLF methods fail to fully consider the unique multivariate coupling mechanism of HPVs and the data heterogeneity under extreme navigation conditions. Existing PSLF methods are mainly divided into three categories: traditional models (TM), shallow learning models (SL), and deep learning (DL). However, traditional models struggle to characterize the complex multi-energy flow coupling characteristics of power systems, shallow learning models have limited adaptive extraction capabilities for time-series features, and single deep learning models offer limited performance improvements. Furthermore, existing combined forecasting methods are only optimized for specific scenarios and are not applicable to HPV operating conditions.

[0003] Patent CN 115392340 A discloses a power load forecasting system for multi-energy electric propulsion vessels. The system comprises: a load information acquisition module, a climate information acquisition module, an energy management global optimization strategy formulation module, a ship control center display device, a wireless local area network (WLAN) module, and a big data cloud platform. The load information acquisition module and the climate information acquisition module transmit historical load data from the multi-energy electric propulsion vessel's operation and marine climate data from the areas it traverses to the big data cloud platform via the WLAN module. The energy management global optimization strategy formulation module is connected to the big data cloud platform via the WLAN module and is used to formulate hourly navigation tasks and average speeds for the multi-energy electric propulsion vessel during future voyages. The big data cloud platform is connected to the WLAN module and is used to receive data uploaded via the WLAN, construct a training sample set, perform regression modeling for the prediction model, and finally output the predicted power load value for the time to be predicted. This existing technology has the following drawbacks: multi-source data and physical information rules cannot be integrated, and multi-scale chaos exists in the complex operating environment of HPV (Hyper-Vehicle Propulsion). These are the shortcomings of the existing technology.

[0004] In view of this, it is very necessary to provide a method, system, device and storage medium for predicting probabilistic loads in power systems to solve the above-mentioned defects in the prior art. Summary of the Invention

[0005] Aiming at the technical problems existing in the prior art that multi-source data and physical information rules cannot be integrated, and there is multi-scale chaos in the complex operation environment of HPV, the present invention provides a probabilistic load forecasting method, system, device, and storage medium for a power system to solve the above technical problems.

[0006] In the first aspect, the present invention provides a probabilistic load forecasting method for a power system, including:

[0007] Step S1: A step of multi-source data collection and preprocessing, which is used to collect multi-source data and perform variable screening on the multi-source data;

[0008] The multi-source data includes the operation records of a hydrogen energy ship system and the waterway hydrological and meteorological monitoring data;

[0009] MIC is used to analyze the correlation between multi-source data for variable screening of multi-source data, and strongly correlated data is selected as input data;

[0010] Among them, MIC is the maximum information coefficient, and its calculation process is as follows:

[0011] Step S11: Divide the multi-source data into a×b grids to obtain an approximate probability density distribution. Under the same a×b grid size, the mathematical expression of the maximum mutual information in all possible grids is: [[ID=二十]]Where x and y represent two random variables, and i represents the i-th grid;

[0012] Step S12: Normalize the maximum mutual information value to the interval [0,1], and the mathematical expression is: ;

[0013] Step S13: Calculate the normalized maximum mutual information value of the grids with a×b < B(N), and select the maximum mutual information value at different scales as the MIC value. Among them, B(N) is a function of the sample quantity, and its numerical size is the 0.6th power of the data volume. The calculation mathematical expression is: .

[0014] Step S2: A step of multi-sequence phase space reconstruction, which is used to reconstruct the multi-source data screened after preprocessing into a unified spatio-temporal feature matrix;

[0015] Multi-sequence phase space reconstruction specifically includes converting the original input sequence into a three-dimensional tensor form by selecting an appropriate embedding dimension d:

[0016] Step S21: Denote that the time series of each observed variable in the original input contains n data points, for T observed variables , Let Y represent the time series of the i-th variable, k=1,2,…,n represent the data point indices of the univariate time series, i=1,2,…,T. Define a multivariate time series Y: ;

[0017] Step S22: Reconstruct the phase space of the multivariate time series Y to obtain preliminary reconstruction points. :

[0018]

[0019] in, and These are the i-th time series. The embedding dimension and time delay, where N is the total number of state points in the reconstructed phase space;

[0020] Step S23: In order to represent the overall state of the system in a high-dimensional phase space, the matrix is... All elements are concatenated in row-major order to form a row vector representing a state point in a high-dimensional phase space, as shown below:

[0021]

[0022] in, , , The total dimension D of the reconstructed phase space is the sum of the embedding dimensions of all involved time series. ;

[0023] Step S24: Obtain the set of reconstructed vectors composed of high-dimensional state vectors. , which constitute the reconstructing attractor of a multivariable system.

[0024] Step S3: The step of constructing the LDE chaotic model. Based on LDE, construct the PSLF electrical parameter and environmental parameter chaotic models respectively, which serve as the core physical constraint terms of the loss function of the BERT-PINN multi-step probabilistic PSLF framework.

[0025] In constructing an LDE chaotic model, training is used to enable... The dynamic equation of the LDE is expressed as: ;

[0026] Step S31: Define the physical residual function based on the dynamic equation: Where F is the right-hand side function of the LDE. For system parameters, automatic differential calculation is used. The time derivative;

[0027] Step S32: Construct the PINN total loss function, which includes a data fitting term. and physical constraints Physical constraints include electrical parameters and environmental parameters The total loss function is expressed as: , ;

[0028] Data fitting term The mathematical expression used to fit observed or experimental data is: , Here are the observed data, and N is the predicted number of points. For a point in time;

[0029] Physical constraints The mathematical expression used to check the physical plausibility of the prediction results is: Where M is the total number of discrete sample points used to calculate the residuals of the equation in the physical loss term, and n is the index of the data points in the univariate time series. Let i be the residual at point i. For input parameters, For a point in time.

[0030] Step S4: The step of constructing the BERT-PINN multi-step probabilistic PSLF framework, which is to build a multi-layer BERT encoder with SA mechanism, embed physical constraint terms, and implement two-stage training for multi-step probabilistic load prediction of hydrogen-powered ship power systems.

[0031] In building a multi-layer BERT encoder with SA mechanism, for the self-attention part, the input sequence... By using a trainable linear transformation matrix , , The multiplication transformation results in a query vector Q, a key vector K, and a value vector V, along with attention weights. Represented as: ;

[0032] Where n is the length of the time series and d is the dimension of the input vector. Given the dimension of the key vector or query vector, a multi-head attention mechanism computes the attention weights of h self-attention heads in parallel. The results are then concatenated to obtain matrix C, which is then processed using a trainable weight matrix. The multi-head attention output is obtained by restoring the original dimension through linear transformation. Matrix C is represented as: ;

[0033] Multi-head attention output is represented as: ;

[0034] A multi-layer BERT encoder with SA mechanism is built. The prediction results are generated by the BERT encoder and then output as a 2×k prediction interval array through a fully connected layer.

[0035] The evaluation metrics for the prediction interval are PICP and PINAW. PICP represents the probability of coverage of the prediction interval, which is used to measure the proportion of actual values ​​falling into the prediction interval and to evaluate the reliability of the model.

[0036] PINAW stands for Normalized Mean Width, which is used to normalize the width of the prediction interval and evaluate the prediction accuracy.

[0037] The mathematical expression for PICP is: ;

[0038] The mathematical expression for PINAW is: ;

[0039] in, This indicates the number of samples in the input data. and Let represent the upper and lower bounds of the i-th prediction interval, respectively. It has two possible values, 0 and 1. If the value falls within the prediction interval, assign a value of 1; otherwise, assign a value of 0. and This indicates the upper and lower bounds of the target value over the entire forecast period;

[0040] CWC is used to combine PICP and PINAW to evaluate the prediction interval. A good prediction interval has a high PICP and a low PINAW. CWC is the coverage width criterion, and its mathematical expression is: ;

[0041] Where η is used as the penalty coefficient for PICPs that fail to meet the target, and μ represents the probability of a PIPC meeting a preset target.

[0042] The two-stage training program includes a pre-training phase and a fine-tuning phase.

[0043] In the pre-training phase, a masking mechanism is introduced into the training to mask some features. Through learning from a large amount of data, the specific features of the masked parts are correctly judged, and the temporal feature relationships between data are learned at the same time.

[0044] The fine-tuning phase learns the relationships between all input features during the pre-training phase and then fine-tunes the parameters for specific downstream tasks.

[0045] Secondly, the technical solution of the present invention also provides a power system probabilistic load prediction system, including a multi-source data acquisition and preprocessing module, a multi-sequence phase space reconstruction module, an LDE chaotic model construction module, and a BERT-PINN multi-step probabilistic PSLF framework construction module;

[0046] The multi-source data acquisition and preprocessing module is used to acquire multi-source data and perform variable filtering on the multi-source data;

[0047] The multi-sequence phase space reconstruction module is used to reconstruct the preprocessed and filtered multi-source data into a unified spatiotemporal feature matrix;

[0048] The LDE chaotic model construction module constructs chaotic models of PSLF electrical parameters and environmental parameters based on LDE, which serve as the core physical constraint terms of the loss function of the BERT-PINN multi-step probabilistic PSLF framework.

[0049] The BERT-PINN multi-step probabilistic PSLF framework building module is used to perform multi-step probabilistic load prediction for hydrogen-powered ship power systems by building a multi-layer BERT encoder with SA mechanism, embedding physical constraint terms, and implementing two-stage training.

[0050] Thirdly, the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor; when the processor executes the program, it implements the power system probabilistic load forecasting method as described in the first aspect.

[0051] Fourthly, the present invention also provides a computer-readable storage medium, wherein the computer program, when executed by a processor, implements the power system probabilistic load forecasting method as described in the first aspect.

[0052] The beneficial effects of this invention are as follows: The method, system, device, and storage medium for probabilistic load prediction of power systems provided by this invention capture chaotic coupling characteristics through a bidirectional multi-head attention mechanism, and introduce chaotic physical constraints based on the Lorentz dynamic equation (LDE) into the loss function. This effectively integrates the data-driven model and physical information rules of multi-source data, and uses multi-sequence phase space reconstruction (MSPSR) to transform ship operation data and waterway environmental monitoring data into a unified spatiotemporal feature matrix. This effectively solves the feature fusion problem of multi-scale chaotic characteristics under the complex operating environment of HPV, and constructs LDE chaotic models for electrical parameters and environmental parameters respectively, which serve as the core physical constraint terms of the loss function, thereby improving the model prediction performance.

[0053] Furthermore, the design principle of this invention is reliable, the structure is simple, and it has a very wide range of application prospects. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart of a probabilistic load prediction method for power systems provided by the present invention.

[0056] Figure 2 This is a principle block diagram of a power system probabilistic load prediction system provided by the present invention. Detailed Implementation

[0057] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0058] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0059] Example 1:

[0060] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for probabilistic load forecasting of a power system, comprising the following steps:

[0061] Step S1: Multi-source data acquisition and preprocessing step, used to acquire multi-source data and perform variable filtering on the multi-source data;

[0062] Multi-source data includes operating records of hydrogen-powered ship systems and hydrological and meteorological monitoring data of waterways. Due to the obvious random fluctuations and intermittent nature of multi-source data, MIC analysis is used to analyze the correlation between multi-source data, filter variables in the multi-source data, and select strongly correlated data as input data.

[0063] Wherein, MIC is the maximum information coefficient, and its calculation process is as follows:

[0064] Step S11: Divide the multi-source data into an a×b grid to obtain an approximate probability density distribution. Under the same a×b grid size, the mathematical expression for the maximum mutual information among all possible grids is: , where x and y represent two random variables, and i represents the i-th grid;

[0065] Step S12: Normalize the maximum mutual information value to the interval [0, 1]. The mathematical expression is: ;

[0066] Step S13: Calculate the normalized maximum mutual information value of the grids where a×b < B(N), and select the maximum mutual information value at different scales as the MIC value. Here, B(N) is a function of the sample quantity, and its numerical value is the 0.6th power of the data volume. The calculation mathematical expression is: .

[0067] Step S2: Steps for multi-sequence phase space reconstruction, which are used to reconstruct the preprocessed and screened multi-source data into a unified spatio-temporal feature matrix;

[0068] The multi-sequence phase space reconstruction converts the original input sequence into a three-dimensional tensor form by selecting an appropriate embedding dimension d, which specifically includes:

[0069] Step S21: Denote that the time series of each observed variable in the original input contains n data points. For T observed variables , represents the time series of the i-th variable, k = 1, 2,..., n represents the data point index of the univariate time series, and i = 1, 2,..., T. Define the multivariate time series Y: ;

[0070] Step S22: Perform phase space reconstruction on the multivariate time series Y to obtain preliminary reconstructed points :

[0071]

[0072] Among them, and are the embedding dimension and time delay of the i-th time series respectively, and N is the total number of state points in the reconstructed phase space;

[0073] Step S23: In order to represent the overall state of the system in the high-dimensional phase space, connect all the elements of the matrix in row order to form a row vector representing a state point in the high-dimensional phase space, as shown below:

[0074]

[0075] Among them, , , The total dimension D of the reconstructed phase space is the sum of the embedding dimensions of all involved time series. ;

[0076] Step S24: Obtain the set of reconstructed vectors composed of high-dimensional state vectors. , which constitute the reconstructing attractor of a multivariable system.

[0077] Step S3: The step of constructing the LDE chaotic model. Based on LDE, construct the PSLF electrical parameter and environmental parameter chaotic models respectively, which serve as the core physical constraint terms of the loss function of the BERT-PINN multi-step probabilistic PSLF framework.

[0078] In constructing an LDE chaotic model, training is used to enable... The dynamic equation of the LDE is expressed as: ;

[0079] Step S31: Define the physical residual function based on the dynamic equation: Where F is the right-hand side function of the LDE. For system parameters, automatic differential calculation is used. The time derivative;

[0080] Step S32: Construct the PINN total loss function, which includes a data fitting term. and physical constraints Physical constraints include electrical parameters and environmental parameters The total loss function is expressed as: , ;

[0081] Data fitting term The mathematical expression used to fit observed or experimental data is: , Here are the observed data, and N is the predicted number of points. For a point in time;

[0082] Physical constraints Electrical parameters used to verify the physical plausibility of prediction results and environmental parameters It can be calculated using the following formula, the mathematical expression of which is: Where M is the total number of discrete sample points used to calculate the residuals of the equation in the physical loss term, and n is the index of the data points in the univariate time series. Let i be the residual at point i. For input parameters, For a point in time.

[0083] Step S4: The step of constructing the BERT-PINN multi-step probabilistic PSLF framework, which is to build a multi-layer BERT encoder with SA mechanism, embed physical constraint terms, and implement two-stage training for multi-step probabilistic load prediction of hydrogen-powered ship power systems.

[0084] In building a multi-layer BERT encoder with SA mechanism, for the self-attention part, the input sequence... By using a trainable linear transformation matrix , , The multiplication transformation results in a query vector Q, a key vector K, and a value vector V, along with attention weights. Represented as: ;

[0085] Where n is the length of the time series and d is the dimension of the input vector. Given the dimension of the key vector or query vector, a multi-head attention mechanism computes the attention weights of h self-attention heads in parallel. The results are then concatenated to obtain matrix C, which is then processed using a trainable weight matrix. The multi-head attention output is obtained by restoring the original dimension through linear transformation. Matrix C is represented as: ;

[0086] Multi-head attention output is represented as: ;

[0087] A multi-layer BERT encoder with SA mechanism is built. The prediction results are generated by the BERT encoder and then output as a 2×k prediction interval array through a fully connected layer.

[0088] The evaluation metrics for the prediction interval are PICP and PINAW. PICP represents the probability of coverage of the prediction interval, which is used to measure the proportion of actual values ​​falling into the prediction interval and to evaluate the reliability of the model.

[0089] PINAW stands for Normalized Mean Width, which is used to normalize the width of the prediction interval and evaluate the prediction accuracy.

[0090] The mathematical expression for PICP is: ;

[0091] The mathematical expression for PINAW is: ;

[0092] in, This indicates the number of samples in the input data. and Let represent the upper and lower bounds of the i-th prediction interval, respectively. It has two possible values, 0 and 1. If the value falls within the prediction interval, assign a value of 1; otherwise, assign a value of 0. and This indicates the upper and lower bounds of the target value over the entire forecast period;

[0093] CWC is used to combine PICP and PINAW to evaluate the prediction interval. A good prediction interval has a high PICP and a low PINAW. CWC is the coverage width criterion, and its mathematical expression is: ;

[0094] Where η is used as the penalty coefficient for PICPs that fail to meet the target, and μ represents the probability of a PIPC meeting a preset target.

[0095] The two-stage training program includes a pre-training phase and a fine-tuning phase.

[0096] In the pre-training phase, a masking mechanism is introduced into the training to mask some features. Through learning from a large amount of data, the specific features of the masked parts are correctly judged, and the temporal feature relationships between data are learned at the same time.

[0097] The fine-tuning phase learns the relationships between all input features during the pre-training phase and then fine-tunes the parameters for specific downstream tasks.

[0098] Example 2:

[0099] like Figure 2 As shown, this embodiment also provides a power system probabilistic load forecasting system, including a multi-source data acquisition and preprocessing module 1, a multi-sequence phase space reconstruction module 2, an LDE chaotic model construction module 3, and a BERT-PINN multi-step probabilistic PSLF framework construction module 4;

[0100] Multi-source data acquisition and preprocessing module 1 is used to acquire multi-source data and perform variable filtering on the multi-source data;

[0101] Multi-sequence phase space reconstruction module 2 is used to reconstruct the preprocessed and filtered multi-source data into a unified spatiotemporal feature matrix;

[0102] LDE Chaotic Model Construction Module 3 constructs chaotic models of PSLF electrical parameters and environmental parameters based on LDE, which serve as the core physical constraint terms of the loss function of the BERT-PINN multi-step probabilistic PSLF framework.

[0103] Module 4 of the BERT-PINN multi-step probabilistic PSLF framework is used to perform multi-step probabilistic load prediction for hydrogen-powered ship power systems by building a multi-layer BERT encoder with SA mechanism, embedding physical constraint terms, and implementing two-stage training.

[0104] Example 3:

[0105] This invention also provides a computer device, which may include a processor, a communication interface, a memory, and a bus, wherein the processor, the communication interface, and the memory communicate with each other via the bus. The bus can be used for information transmission between electronic devices and sensors. The processor can call logical instructions in the memory to execute the following method: Step S1: a multi-source data acquisition and preprocessing step, used to acquire multi-source data and perform variable filtering on the multi-source data;

[0106] Step S2: The multi-sequence phase space reconstruction step, used to reconstruct the preprocessed and filtered multi-source data into a unified spatiotemporal feature matrix;

[0107] Step S3: The step of constructing the LDE chaotic model. Based on LDE, construct the PSLF electrical parameter and environmental parameter chaotic models respectively, which serve as the core physical constraint terms of the loss function of the BERT-PINN multi-step probabilistic PSLF framework.

[0108] Step S4: The step of constructing the BERT-PINN multi-step probabilistic PSLF framework is to build a multi-layer BERT encoder with SA mechanism, embed physical constraint terms, and implement two-stage training to perform multi-step probabilistic load prediction for hydrogen-powered ship power systems.

[0109] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0110] Example 4:

[0111] This invention provides a computer-readable storage medium storing computer instructions that cause a computer to execute the method provided in the above-described method embodiments, such as: step S1: a multi-source data acquisition and preprocessing step, used to acquire multi-source data and perform variable filtering on the multi-source data;

[0112] Step S2: The multi-sequence phase space reconstruction step, used to reconstruct the preprocessed and filtered multi-source data into a unified spatiotemporal feature matrix;

[0113] Step S3: The step of constructing the LDE chaotic model. Based on LDE, construct the PSLF electrical parameter and environmental parameter chaotic models respectively, which serve as the core physical constraint terms of the loss function of the BERT-PINN multi-step probabilistic PSLF framework.

[0114] Step S4: The step of constructing the BERT-PINN multi-step probabilistic PSLF framework is to build a multi-layer BERT encoder with SA mechanism, embed physical constraint terms, and implement two-stage training to perform multi-step probabilistic load prediction for hydrogen-powered ship power systems.

[0115] In this technical solution, PSLF refers to Power System Load Forecasting, BERT-PINN refers to the PINN architecture built by the bidirectional encoder Transformer. The Transformer model is a deep neural network, PINN is a physical information neural network, BERT is a bidirectional encoder representation based on Transformer, and SA refers to the attention mechanism.

[0116] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The methods disclosed in the embodiments are described simply because they correspond to the systems disclosed in the embodiments; relevant details can be found in the method section.

[0117] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0118] In the embodiments provided by this invention, it should be understood that the disclosed systems, methods, and approaches can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0119] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0120] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit.

[0121] Similarly, in the various embodiments of the present invention, each processing unit can be integrated into a functional module, or each processing unit can exist physically, or two or more processing units can be integrated into a functional module.

[0122] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0123] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0124] The above-disclosed embodiments are merely preferred embodiments of the present invention, but the present invention is not limited thereto. Any non-creative variations that can be conceived by those skilled in the art, as well as any improvements and modifications made without departing from the principles of the present invention, should fall within the protection scope of the present invention.

Claims

1. A method for probabilistic load forecasting in a power system, characterized in that, Includes the following steps: Step S1: Multi-source data acquisition and preprocessing step, used to acquire multi-source data and perform variable filtering on the multi-source data; Step S2: The multi-sequence phase space reconstruction step, used to reconstruct the preprocessed and filtered multi-source data into a unified spatiotemporal feature matrix; Step S3: The step of constructing the LDE chaotic model. Based on LDE, construct the PSLF electrical parameter and environmental parameter chaotic models respectively, which serve as the core physical constraint terms of the loss function of the BERT-PINN multi-step probabilistic PSLF framework. Define the dynamic equation of the LDE, and its mathematical expression is: Through training ; Step S31: Define the physical residual function based on the LDE dynamic equations: Where F is the right-hand side function of the LDE. For system parameters, automatic differential calculation is used. The time derivative; Step S32: Construct the PINN total loss function, which includes a data fitting term. and physical constraints Physical constraints include electrical parameters and environmental parameters The total loss function is expressed as: , ; Data fitting term The mathematical expression used to fit observed or experimental data is: , Here are the observed data, and N is the predicted number of points. For a point in time; Physical constraints The mathematical expression used to check the physical plausibility of the prediction results is: Where M is the total number of discrete sample points used to calculate the residuals of the equation in the physical loss term, and n is the index of the data points in the univariate time series. Let i be the residual at point i. For input parameters, For a point in time; Step S4: The step of constructing the BERT-PINN multi-step probabilistic PSLF framework is to build a multi-layer BERT encoder with SA mechanism, embed physical constraint terms, and implement two-stage training to perform multi-step probabilistic load prediction for hydrogen-powered ship power systems.

2. The power system probabilistic load forecasting method according to claim 1, characterized in that, The multi-source data includes hydrogen-powered ship system operation records and waterway hydrological and meteorological monitoring data; The variable screening of multi-source data uses MIC analysis to analyze the correlation between multi-source data and selects strongly correlated data as input data. Wherein, MIC is the maximum information coefficient, and its calculation process is as follows: Step S11: Divide the multi-source data into an a×b grid to obtain an approximate probability density distribution. Under the same a×b grid size, the mathematical expression for the maximum mutual information among all possible grids is: Where x and y represent two random variables, and i represents the i-th grid; Step S12: Normalize the maximum mutual information value to the interval [0,1]. The mathematical expression is: ; Step S13: Calculate the normalized maximum mutual information value for the grid with a×b < B(N), and select the maximum mutual information value at different scales as the MIC value, where B(N) is a function of the number of samples, and its value is 0.6 times the data size. The mathematical expression for calculation is: .

3. The power system probabilistic load forecasting method according to claim 1, characterized in that, The multi-sequence phase space reconstruction, by selecting an appropriate embedding dimension d, converts the original input sequence into a three-dimensional tensor form, specifically including: Step S21: Let the time series of each observed variable in the original input contain n data points, and for T observed variables... , Let Y represent the time series of the i-th variable, k=1,2,…,n represent the data point indices of the univariate time series, i=1,2,…,T. Define a multivariate time series Y: ; Step S22: Reconstruct the phase space of the multivariate time series Y to obtain preliminary reconstruction points. : in, and These are the i-th time series. The embedding dimension and time delay, where N is the total number of state points in the reconstructed phase space; Step S23: In order to represent the overall state of the system in a high-dimensional phase space, the matrix is... All elements are concatenated in row-major order to form a row vector representing a state point in a high-dimensional phase space, as shown below: in, , , The total dimension D of the reconstructed phase space is the sum of the embedding dimensions of all involved time series. ; Step S24: Obtain the set of reconstructed vectors composed of high-dimensional state vectors. , which constitute the reconstructing attractor of a multivariable system.

4. The power system probabilistic load forecasting method according to claim 1, characterized in that, In the construction of the multi-layer BERT encoder with SA mechanism, for the self-attention part, the input sequence By using a trainable linear transformation matrix , , The multiplication transformation results in a query vector Q, a key vector K, and a value vector V, along with attention weights. Represented as: ; Where n is the length of the time series and d is the dimension of the input vector. Given the dimension of the key vector or query vector, a multi-head attention mechanism computes the attention weights of h self-attention heads in parallel. The results are then concatenated to obtain matrix C, which is then processed using a trainable weight matrix. The multi-head attention output is obtained by restoring the original dimension through linear transformation. Matrix C is represented as: ; Multi-head attention output is represented as: .

5. The power system probabilistic load forecasting method according to claim 4, characterized in that, The process involves constructing a multi-layer BERT encoder with SA mechanism, generating prediction results through the BERT encoder, and then outputting a 2×k prediction interval array through a fully connected layer. The evaluation metrics for the prediction interval are PICP and PINAW, where PICP represents the probability of coverage of the prediction interval. PINAW stands for Normalized Mean Width, which is used to normalize the width of the prediction interval. The mathematical expression for PICP is: ; The mathematical expression for PINAW is: ; in, This indicates the number of samples in the input data. and Let represent the upper and lower bounds of the i-th prediction interval, respectively. It has two possible values, 0 and 1. If the value falls within the prediction interval, assign a value of 1; otherwise, assign a value of 0. and This indicates the upper and lower bounds of the target value over the entire forecast period; CWC is used to combine PICP and PINAW to evaluate the prediction interval. A good prediction interval has a high PICP and a low PINAW. CWC is the coverage width criterion, and its mathematical expression is: ; Where η is used as the penalty coefficient for PICPs that fail to meet the standard, and μ represents the preset target probability that PIPCs must meet.

6. The power system probabilistic load forecasting method according to claim 1, characterized in that, The two-stage training includes a pre-training stage and a fine-tuning stage. In the pre-training phase, a masking mechanism is introduced into the training to mask some features. Through learning from a large amount of data, the specific features of the masked parts are correctly judged, and the temporal feature relationships between data are learned at the same time. The fine-tuning phase learns the relationships between all input features during the pre-training phase and then fine-tunes the parameters for specific downstream tasks.

7. A power system probabilistic load forecasting system, characterized in that, It includes a multi-source data acquisition and preprocessing module, a multi-sequence phase space reconstruction module, an LDE chaotic model construction module, and a BERT-PINN multi-step probabilistic PSLF framework construction module; The multi-source data acquisition and preprocessing module is used to acquire multi-source data and perform variable filtering on the multi-source data; The multi-sequence phase space reconstruction module is used to reconstruct the preprocessed and filtered multi-source data into a unified spatiotemporal feature matrix; The LDE chaotic model construction module constructs chaotic models of PSLF electrical parameters and environmental parameters based on LDE, which serve as the core physical constraint terms of the loss function of the BERT-PINN multi-step probabilistic PSLF framework. Define the dynamic equation of the LDE, and its mathematical expression is: Through training ; Step S31: Define the physical residual function based on the LDE dynamic equations: Where F is the right-hand side function of the LDE. For system parameters, automatic differential calculation is used. The time derivative; Step S32: Construct the PINN total loss function, which includes a data fitting term. and physical constraints Physical constraints include electrical parameters and environmental parameters The total loss function is expressed as: , ; Data fitting term The mathematical expression used to fit observed or experimental data is: , Here are the observed data, and N is the predicted number of points. For a point in time; Physical constraints The mathematical expression used to check the physical plausibility of the prediction results is: Where M is the total number of discrete sample points used to calculate the residuals of the equation in the physical loss term, and n is the index of the data points in the univariate time series. Let i be the residual at point i. For input parameters, For a point in time; The BERT-PINN multi-step probabilistic PSLF framework building module is used to perform multi-step probabilistic load prediction for hydrogen-powered ship power systems by building a multi-layer BERT encoder with SA mechanism, embedding physical constraint terms, and implementing two-stage training.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the power system probabilistic load forecasting method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the power system probabilistic load forecasting method as described in any one of claims 1-6.

Images