Hybrid power generation system energy prediction method based on multi-field unified modeling

Abstract

The invention discloses a hybrid power generation system energy prediction method based on multi-field unified modeling, and belongs to the technical field of power system energy control, and the method comprises the steps: selecting system working parameters, and constructing constraint conditions; constructing dynamic working models of the power generation system, wherein the dynamic working models comprise a hydrogen fuel cell working model and a lithium battery dynamic model; deploying a WSN module to collect working parameters of key parts of the hybrid power generation system in real time, and establishing a standard data set by adopting a dynamic filtering algorithm; classifying the standard data set into a corresponding cluster by adopting a clustering space input algorithm; and establishing a fuzzy prediction model based on a four-layer neural network, extracting related data from the standard data set according to a membership function, taking the related data as input of the fuzzy prediction model, and obtaining an energy prediction value. According to the method, adaptive energy prediction is realized through updating iteration, the energy distribution efficiency is improved, and accurate prediction of multi-field energy in the hybrid power generation system can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of power system energy control, and in particular to a hybrid power generation system energy prediction method based on multi-domain unified modeling. Background Art

[0002] Hybrid power generation systems integrate multiple energy sources, including energy storage, batteries, and hydrogen, aiming to achieve a more efficient, reliable, and sustainable energy supply through energy complementarity and optimized scheduling. However, traditional energy control methods typically rely on adaptive algorithms for direct energy allocation. These methods present significant challenges in building multimodal prediction models, suffer from poor real-time performance, and struggle to effectively address complex issues in real-world applications.

[0003] Existing energy forecasting algorithms for hybrid power systems are primarily based on statistics, machine learning, deep learning, and the combination of physical models and optimization algorithms. Statistical time series analysis (such as ARIMA and SARIMA) uses historical data to predict future energy output. While simple and easy to implement, these methods are ineffective in handling nonlinearities and complex fluctuations, and are not adaptable to extreme weather or uncertainties. Machine learning-based regression methods (such as support vector regression (SVR) and random forest (RF)) can better address nonlinearities by mining the relationship between historical data and power generation. However, they also fail to fully account for the dynamic complementarity between multiple energy sources and require significant computational resources and training data. Furthermore, deep learning methods, particularly long-short-term memory (LSTM) networks and convolutional neural networks (CNNs), and their combinations, can capture long-term and short-term dependencies and spatial features in time series data, significantly improving forecast accuracy. However, these methods require large amounts of data and computational resources and still suffer from latency issues in real-time performance. Methods that combine physical and statistical models, typically combining meteorological forecasts with physical models, can provide relatively accurate energy forecasts. However, these methods rely on precise physical parameters and assumptions, making them difficult to adapt to complex dynamic environments. Optimization algorithms (such as genetic algorithm GA, particle swarm optimization PSO, etc.) have good applications in scheduling optimization, but they focus on scheduling decisions and lack accurate predictions of energy production, energy storage dynamics, etc., which affects the overall system performance.

[0004] The existing technical solutions for hybrid power generation system energy management have the following technical problems:

[0005] Many existing energy forecasting methods only model a single type of renewable energy (such as wind power or photovoltaics), ignoring the interactions and complementarities between different energy sources. In actual hybrid power generation systems, these methods fail to fully exploit the advantages of multiple energy sources, resulting in reduced forecasting accuracy. For example, the generation characteristics of wind and solar power are complementary to a certain extent, but existing methods often fail to effectively incorporate this complementarity into forecasting models, resulting in an inability to accurately grasp the overall system's operating status in short-term forecasts. Furthermore, in hybrid power generation systems, the complementarity between different energy sources is crucial. For example, lithium batteries and fuel cells often have different generation times and intensities, but they can complement each other. However, current forecasting models often only forecast each energy source separately, lacking unified modeling of complementary characteristics. As a result, forecast results fail to reflect the coordinated effects of energy sources.

[0006] Furthermore, because hybrid power generation systems include multiple energy sources and storage systems, existing forecasting algorithms typically require processing a large number of variables and complex mathematical models, resulting in a high computational load. Traditional models are often difficult to update in real time, resulting in delayed forecast results and an inability to meet the real-time scheduling requirements of power systems.

[0007] Chinese invention patent publication number CN114079284B discloses a method for energy management and optimization control of a distributed power generation system that takes into account the efficiency characteristics of fuel cells. For distributed power generation systems, energy flow regulation is performed using an energy optimization approach. Chinese utility model patent publication number CN204391744U discloses an energy controller for a multi-energy input power generation system that achieves unified energy distribution in multiple fields.

[0008] Existing methods still lack comprehensive modeling of the complementarities between different energy sources, and there is still significant room for improvement in terms of real-time performance, computational efficiency, and prediction accuracy. This is especially true when dealing with multi-domain interactions and complex dynamic changes, making it difficult to provide accurate and timely scheduling support. Therefore, developing an energy prediction algorithm for hybrid power generation systems that integrates multi-domain modeling is crucial for improving renewable energy prediction accuracy, optimizing grid scheduling, and ensuring system stability. Summary of the Invention

[0009] The problem to be solved by the present invention is to provide an energy prediction method for a hybrid power generation system based on multi-domain unified modeling, realize adaptive energy prediction through updating and iteration, improve energy distribution efficiency, and realize efficient multi-modal energy distribution in the future.

[0010] The present invention adopts the following technical solution: a hybrid power generation system energy prediction method based on multi-domain unified modeling, comprising the following steps:

[0011] Step 1: Select the operating parameters of the hybrid power generation system and establish the constraints;

[0012] Step 2: Using the operating parameters of the hybrid power generation system, a dynamic operating model of the power generation system is constructed, including a hydrogen fuel cell operating model and a lithium battery dynamic model.

[0013] Step 3: Deploy the WSN module to collect the working parameters of the key parts of the hybrid power generation system in real time and use the dynamic filtering algorithm to establish a standard data set;

[0014] Step 4: Use the clustering space input algorithm to classify the standard data set into the corresponding clusters;

[0015] Step 5: Establish a fuzzy prediction model based on a four-layer neural network, extract relevant data from the standard data set according to the membership function, and use it as the input of the fuzzy prediction model to obtain the energy prediction value.

[0016] Preferably, in step 1, the operating parameters of the hybrid power generation system include: hydrogen fuel cell output current, hydrogen fuel internal resistance, lithium battery output current, lithium battery internal resistance, and lithium battery SOC.

[0017] Preferably, in step 2, the hydrogen fuel cell working model is expressed as:

[0018] V FC =E OCV -iR FC -η FC i

[0019] Among them, V FC is the output voltage of the fuel cell, E OCV is the open circuit voltage, i is the battery output current, R FC is the internal resistance of the battery, η FC is the efficiency coefficient;

[0020] The dynamic model of lithium battery is expressed as:

[0021] V LIB =E OCV (SOC)-iR LIB -η LIB i

[0022] Among them, V LIB is the output voltage of the lithium battery, SOC is the remaining capacity of the battery, R LIB is the internal resistance of the lithium battery, η LIB is the charge and discharge efficiency.

[0023] Preferably, in step 3, a dynamic filtering algorithm is used to predict the battery voltage based on the measured value to estimate the battery state;

[0024] The battery prediction status is expressed as follows:

[0025]

[0026] Among them, the state variable is the battery voltage at the current moment k; u k-1 is the charge and discharge current of the battery; A is the state transfer matrix; B is the control input matrix;

[0027] The relationship between the measured value and the predicted value is expressed as follows:

[0028]

[0029] Among them, y k Indicates the battery voltage V measured at the current moment FC and V LIB ; H is the observation matrix, which represents the relationship between battery voltage, SOC and current; ω k represents the measurement noise;

[0030] The predicted error covariance matrix is obtained according to the state transfer matrix, which is expressed as:

[0031]

[0032] Among them, P k-1 is the error covariance matrix at the previous moment, represents the predicted covariance matrix, Q is the process noise covariance matrix, which represents the random noise or uncertainty of the model itself, and the superscript T represents the transpose;

[0033] According to the relationship between the predicted value and the actual value, the coefficient between the predicted value and the actual value is obtained. The calculation formula is:

[0034]

[0035] Among them, R is the accuracy of the voltage sensor, which reflects the measurement error value.

[0036] Update the battery state estimate based on the new measurements and predictions:

[0037]

[0038] in, refers to the updated SOC estimate;

[0039] The updated covariance matrix is obtained to describe the accuracy of the current state estimation, which is expressed as:

[0040]

[0041] Where I represents the identity matrix and represents the correction after the update.

[0042] Preferably, in step 4, construct the input feature vector v k As the input of the clustering space input algorithm, it is expressed as:

[0043]

[0044] Among them, diag(P k ) represents the error covariance matrix P k The diagonal elements of represent the estimated uncertainty of each state component.

[0045] Preferably, in step 4, the cluster space input algorithm is based on the feature vector v k Perform cluster analysis, including the following sub-steps:

[0046] Step 4.1, randomly select K initial cluster centers c1, c2, ..., c k , each cluster center is a d-dimensional vector, the same dimension as the input space;

[0047] Step 4.2, calculate each data point v k The Euclidean distance to the cluster center dist(v k ,c i ), expressed as:

[0048] dist(v k ,c i )=||v k -c i ||

[0049] Step 4.3, the data point v k Assign to the nearest cluster center:

[0050] cluster k =argmindist i (v k ,c i )

[0051] Among them, cluster k represents the operating state of the system at time step k;

[0052] Step 4.4: Based on the new cluster assignment, recalculate the center of each cluster and update the cluster center to the mean of all data points in the cluster:

[0053]

[0054] Among them, N i is the number of data points in cluster i;

[0055] Step 4.5, repeat steps 4.1 to 4.4 until the preset maximum number of iterations is completed based on the feature vector v k Cluster analysis of .

[0056] Preferably, in step 5, the membership function is expressed as:

[0057]

[0058] Among them, x i represents the i-th data point, c j represents the center of cluster j, ||x i -c j || represents the data point x i To cluster center c j The Euclidean distance of , m is the fuzzification index.

[0059] Preferably, in step 5, building a fuzzy prediction model based on a four-layer neural network includes the following sub-steps:

[0060] Step 5.1: Build the input layer, expressed as:

[0061]

[0062] in, represents the original input features, u ij is the data point x i The membership degree of cluster j, C represents the number of clusters;

[0063] Step 5.2: Calculate the first layer. The calculation formula is:

[0064] h1=σ(W1x i +b1)

[0065] Where h1 represents the number of neurons in the first layer, b1 represents the bias term of the first layer, σ is the activation function, and W1 represents the weight matrix of the first layer;

[0066] Step 5.3: Calculate the second and third layers using the following formula:

[0067]

[0068] Among them, W2 and W3 represent the weight matrices of the second and third layers, h2 and h3 represent the number of neurons in the second and third layers, and b2 and b3 represent the bias terms of the second and third layers;

[0069] Step 5.4, build the output layer, the formula is:

[0070]

[0071] in, Represents the predicted value, W4 is the weight matrix of the output layer, and b4 is the bias term of the output layer.

[0072] Step 5.4: Use MSE as the loss function to train the neural network.

[0073] The technical solution of the present invention further provides: an electronic device, comprising:

[0074] one or more processors;

[0075] a storage device having one or more programs stored thereon;

[0076] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the above-mentioned hybrid power generation system energy prediction methods based on multi-domain unified modeling.

[0077] The technical solution of the present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, it implements the steps of any of the above-mentioned hybrid power generation system energy prediction methods based on multi-domain unified modeling.

[0078] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0079] 1. The energy forecasting method of this invention, through unified multi-energy modeling, can simultaneously consider the interactions between various energy sources. This helps provide more comprehensive and coordinated forecast results, especially when renewable energy is highly volatile. By jointly modeling the characteristics of different energy sources, the neural network can capture the correlations between them, providing strong support for optimizing overall power generation strategies and scheduling decisions in the future.

[0080] 2. The energy prediction method of this invention eliminates noise and errors in sensor data, making the input data more accurate and reliable. Preprocessing enables the model to be trained on cleaner, more accurate data, improving prediction accuracy. Filtering removes short-term fluctuations in the data while preserving long-term trends, helping the neural network better capture key patterns in energy generation and preventing short-term fluctuations from affecting prediction results.

[0081] 3. The method of the present invention can help segment data through cluster analysis. Different prediction models can be applied to each data segment. Clustering can establish specialized prediction models for different current intervals, thereby improving prediction accuracy. At the same time, the present invention can extract meaningful patterns or categories from large amounts of data and divide them into different power generation states. This helps the model recognize features under different conditions during training and improves the precision of predictions.

[0082] 4. The method of the present invention uses a multi-layer hidden layer structure to automatically extract meaningful features from raw data. The four-layer neural network effectively processes the complex nonlinear relationship between input data and output through multiple layers of nonlinear activation functions, thereby significantly improving prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 This is a flowchart of the energy prediction method for the hybrid power generation system of the present invention;

[0084] Figure 2 This is a diagram of the four-layer feedforward neural network prediction model of the present invention. DETAILED DESCRIPTION

[0085] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the application are further elaborated in detail below with reference to the accompanying drawings. The described embodiments are only a part of the embodiments involved in the present invention. All non-innovative embodiments of other researchers in this field on this embodiment fall within the scope of protection of the present invention. At the same time, the step numbers in the embodiments of the present invention are only set for the convenience of explanation and description, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0086] In one embodiment of the present invention, a hybrid power generation system energy prediction method based on multi-domain unified modeling realizes adaptive energy prediction through updating and iteration, which can improve energy distribution efficiency and realize efficient multi-modal energy distribution in the future. Figure 1 As shown, the specific steps include:

[0087] Step 1: Select the operating parameters of the hybrid power generation system and establish the constraints;

[0088] In this embodiment, the operating parameters of the hybrid power generation system include: hydrogen fuel cell output current, hydrogen fuel internal resistance, lithium battery output current, lithium battery internal resistance, and lithium battery SOC.

[0089] In this embodiment, the constraints are expressed as follows:

[0090]

[0091] Among them, i(t) is the output current of the battery at time t, i rated (t) Battery rated maximum current; P ESS,discharge Indicates the minimum discharge power of the battery, P ESS,charge Indicates the maximum charging power of the battery, P ESS (t) represents the charge and discharge power of the battery at time t; SOC(t) represents the residual current of the battery at time t, SOC minIndicates the minimum SOC allowed by the battery, SOC max Indicates the maximum SOC allowed by the battery; R FC (t) represents the internal resistance of the battery at time t, is the minimum internal resistance of the battery, is the maximum internal resistance of the battery.

[0092] Step 2: Construct a dynamic operating model of the power generation system by mixing the operating parameters of the power generation system.

[0093] In this embodiment, the dynamic working model of the power generation system includes: a hydrogen fuel cell working model and a lithium battery dynamic model, which are respectively used to obtain the output voltage V FC and the output voltage of the lithium battery V LIB .

[0094] Step 3: Deploy WSN and collect data, and use dynamic filtering algorithm to establish a standard data set.

[0095] In this embodiment, the WSN module is used to collect real-time operating parameters of key components of the hybrid power generation system, including hydrogen fuel cell output current, hydrogen fuel internal resistance, lithium battery output current, lithium battery internal resistance, and lithium battery state of charge (SOC). A dynamic filtering algorithm is then used to remove noise before transferring the data to a dataset.

[0096] Specifically, a dynamic filtering algorithm estimates the state of the battery in the battery system based on the system model and measurements, providing a smoother and more accurate voltage prediction.

[0097] The battery prediction status is as follows:

[0098]

[0099] Among them, the state variable is the battery voltage at the current moment k, i.e. the internal state of the battery; u k-1 It refers to the battery charge and discharge current i; A is the state transfer matrix, which is obtained through the dynamic characteristics of the battery and represents the changing relationship between the battery voltage and SOC; B is the control input matrix, which describes how the battery current affects the state of the system.

[0100] At the same time, the relationship between the measured value and the predicted value is given as follows:

[0101]

[0102] Among them, y k Indicates the battery voltage V measured at the current moment FC and V LIB ; H is the observation matrix, which represents the relationship between battery voltage, SOC and current; ω kRepresents measurement noise, which refers to sensor errors or interference from the external environment.

[0103] The predicted error covariance matrix is obtained according to the state transfer matrix, which is expressed as:

[0104]

[0105] Among them, P k-1 is the error covariance matrix at the previous moment, and Q is the process noise covariance matrix, which represents the random noise or uncertainty of the model itself.

[0106] According to the relationship between the predicted value and the actual value, the coefficient between the predicted value and the actual value is obtained. The calculation formula is:

[0107]

[0108] Among them, R is the accuracy of the voltage sensor, which reflects the measurement error value.

[0109] Update the battery state estimate based on the new measurements and predictions:

[0110]

[0111] in, refers to the updated SOC estimate;

[0112] At the same time, the updated covariance matrix is obtained, which describes the accuracy of the current state estimation and is expressed as:

[0113]

[0114] Where I represents the identity matrix and represents the correction after the update.

[0115] Step 4: Use the clustering spatial input algorithm to classify the data set into the corresponding clusters.

[0116] In order to improve the computational efficiency and correctness of the clustering algorithm, this embodiment constructs the input feature vector v k , as the input of clustering, is expressed as:

[0117]

[0118] Among them, diag(P k ) represents the error covariance matrix P k The diagonal elements of represent the estimation uncertainty of each state component; k Indicates the battery charge and discharge power instruction, reflecting the impact of the control strategy on the system; k Battery terminal voltage, plays a role in compensating state variables The role of estimation bias.

[0119] First, randomly select K initial cluster centers c1, c2, ..., c k , each cluster center is a d-dimensional vector (the same dimension as the input space).

[0120] Second, calculate each data point v k The Euclidean distance to the cluster center can be expressed as:

[0121] dist(v k ,c i )=||v k -c i ||

[0122] The data point v k Assign to the nearest cluster center:

[0123] cluster k =argmindist i (v k ,c i )

[0124] Based on the new cluster assignment, recalculate the center of each cluster and update the cluster center to the mean of all data points in the cluster:

[0125]

[0126] Among them, cluster k represents the operating state of the system at time step k, N i is the number of data points in cluster i.

[0127] Finally, the calculation is repeated until the maximum number of iterations is reached, and the eigenvector v is completed. k Cluster analysis of .

[0128] Step 5: Establish a fuzzy prediction model based on a four-layer neural network, extract relevant data from the data set according to the membership function, and use it as the input of the prediction model to obtain the predicted value.

[0129] As a criterion for data extraction, the membership function can be expressed as:

[0130]

[0131] Among them, x i represents the i-th data point, c j represents the center of cluster j, ||x i -c j || represents the data point x i To cluster center c j The Euclidean distance of , m is the fuzzification index.

[0132] Preferably, in this embodiment, m is selected to be 2.

[0133] In order to improve the nonlinear ability and adaptability of the algorithm, this embodiment establishes a fuzzy prediction model based on a four-layer neural network, such as Figure 2 As shown, the input layer is represented as:

[0134]

[0135] in, represents the original input characteristics (current, voltage), u ij is the data point x i The degree of membership to cluster j.

[0136] In this implementation, the number of neurons in the input layer is d+C, where d represents the dimension of the original feature and C represents the number of clusters. Therefore, the calculation formula for the first layer is deduced as:

[0137] h1=σ(W1x i +b1)

[0138] Among them, W1 represents the weight matrix of the first layer, the size is h1×(d+C), h1 represents the number of neurons in the first layer, b1 represents the bias term of the first layer, and σ is the activation function.

[0139] Similarly, the formulas for the second and third layers are:

[0140]

[0141] Among them, W2 and W3 represent the weight matrices of the second and third layers, with sizes of h2×h1 and h3×h2, h2 and h3 represent the number of neurons in the second and third layers, and b2 and b3 represent the bias terms of the second and third layers.

[0142] Finally, build the output layer. The core formula is:

[0143]

[0144] in, Represents the predicted value, W4 is the weight matrix of the output layer, size is 1×h3, and b4 is the bias term of the output layer.

[0145] In addition, when training the neural network in this embodiment, MSE is used as the loss function, which is expressed as:

[0146]

[0147] in, is the predicted output of the neural network, y i is the actual output of the hybrid power generation system, and N is the number of data points.

[0148] In an embodiment of the present invention, an electronic device is also provided, including: one or more processors; a storage device on which one or more programs are stored; when the one or more programs are executed by the one or more processors, the one or more processors implement the hybrid power generation system energy prediction method based on multi-domain unified modeling described in the above embodiment.

[0149] In an embodiment of the present invention, a computer-readable storage medium is further provided, on which a computer program is stored. When the program is executed by a processor, the steps of the hybrid power generation system energy prediction method based on multi-domain unified modeling in the above embodiment are implemented.

[0150] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A hybrid power generation system energy prediction method based on multi-domain unified modeling, characterized in that: The steps include: Step 1: Select the operating parameters of the hybrid power generation system and establish the constraints; Step 2: Using the operating parameters of the hybrid power generation system, a dynamic operating model of the power generation system is constructed, including a hydrogen fuel cell operating model and a lithium battery dynamic model. Step 3: Deploy the WSN module to collect the working parameters of the key parts of the hybrid power generation system in real time and use the dynamic filtering algorithm to establish a standard data set; Step 4: Use the clustering space input algorithm to classify the standard data set into the corresponding clusters; Step 5: Establish a fuzzy prediction model based on a four-layer neural network, extract relevant data from the standard data set according to the membership function, and use it as the input of the fuzzy prediction model to obtain the energy prediction value.

2. The hybrid power generation system energy prediction method based on multi-domain unified modeling according to claim 1 is characterized in that: In step 1, the operating parameters of the hybrid power generation system include: hydrogen fuel cell output current, hydrogen fuel internal resistance, lithium battery output current, lithium battery internal resistance, and lithium battery SOC; The constraints are expressed as follows: Among them, i(t) is the output current of the battery at time t, i rated (t) Battery rated maximum current; P ESS,discharge Indicates the minimum discharge power of the battery, P ESS,charge Indicates the maximum charging power of the battery, P ESS (t) represents the charge and discharge power of the battery at time t; SOC(t) represents the residual current of the battery at time t, SOC min Indicates the minimum SOC allowed by the battery, SOC max Indicates the maximum SOC allowed by the battery; R FC (t) represents the internal resistance of the battery at time t, is the minimum internal resistance of the battery, is the maximum internal resistance of the battery.

3. The hybrid power generation system energy prediction method based on multi-domain unified modeling according to claim 1 is characterized in that: In step 2, the hydrogen fuel cell working model is expressed as: V FC =E OCV -iR FC -η FC I Among them, V FC is the output voltage of the fuel cell, E OCV is the open circuit voltage, i is the battery output current, R FC is the internal resistance of the battery, η FC is the efficiency coefficient; The lithium battery dynamic model is expressed as: V LIB =And OCV (SOC)-iR LIB -η LIB i Among them, V LIB is the output voltage of the lithium battery, SOC is the remaining capacity of the battery, R LIB is the internal resistance of the lithium battery, η LIB is the charge and discharge efficiency.

4. The hybrid power generation system energy prediction method based on multi-domain unified modeling according to claim 3 is characterized in that: In step 3, the dynamic filtering algorithm is used to predict the battery voltage based on the measured value and perform battery state estimation; The battery prediction status is expressed as follows: Among them, the state variable is the battery voltage at the current moment k; u k-1 is the charge and discharge current of the battery; A is the state transfer matrix; B is the control input matrix; The relationship between the measured value and the predicted value is expressed as follows: Among them, y k Indicates the battery voltage V measured at the current moment FC and V LIB ; H is the observation matrix, which represents the relationship between battery voltage, SOC and current; ω k represents the measurement noise; The predicted error covariance matrix is obtained according to the state transfer matrix, which is expressed as: P k - =AP k-1 From T +Q Among them, P k-1 is the error covariance matrix of the previous moment, P k - represents the predicted covariance matrix, Q is the process noise covariance matrix, which represents the random noise or uncertainty of the model itself, and the superscript T represents the transpose; According to the relationship between the predicted value and the actual value, the coefficient between the predicted value and the actual value is obtained. The calculation formula is: K k =P k - H T (HP k - H T +R) -1 Among them, R is the accuracy of the voltage sensor, reflecting the measurement error value; Update the battery state estimate based on the new measurements and predictions: in, refers to the updated SOC estimate; The updated covariance matrix is obtained to describe the accuracy of the current state estimation, which is expressed as: P k =(I-K k H)P k - Where I represents the identity matrix and represents the correction after the update.

5. The hybrid power generation system energy prediction method based on multi-domain unified modeling according to claim 4 is characterized in that: In step 4, construct the input feature vector v k As the input of the clustering space input algorithm, it is expressed as: Among them, diag(P k ) represents the error covariance matrix P k The diagonal elements of represent the estimation uncertainty of each state component; k Indicates the battery charge and discharge power instruction, reflecting the impact of the control strategy on the system; k Battery terminal voltage, used to compensate for state variables Estimation bias.

6. The hybrid power generation system energy prediction method based on multi-domain unified modeling according to claim 5 is characterized in that: In step 4, the cluster space input algorithm is based on the feature vector v k Perform cluster analysis, including the following sub-steps: Step 4.1, randomly select K initial cluster centers c1, c2, ..., c k , each cluster center is a d-dimensional vector, the same dimension as the input space; Step 4.2, calculate each data point v k The Euclidean distance to the cluster center dist(v k ,c i ), expressed as: dist(v k ,c i )=||v k -c i || Step 4.3, the data point v k Assign to the nearest cluster center: cluster k =argmindist i (v k ,c i ) Among them, cluster k represents the operating state of the system at time step k; Step 4.4: Based on the new cluster assignment, recalculate the center of each cluster and update the cluster center to the mean of all data points in the cluster: Among them, N i is the number of data points in cluster i; Step 4.5, repeat steps 4.1 to 4.4 until the preset maximum number of iterations is completed based on the feature vector v k Cluster analysis of .

7. The hybrid power generation system energy prediction method based on multi-domain unified modeling according to claim 6 is characterized in that: In step 5, the membership function is expressed as: Among them, x i represents the i-th data point, c j represents the center of cluster j, ||x i -c j || represents the data point x i To cluster center c j The Euclidean distance of , m is the fuzzification index.

8. The hybrid power generation system energy prediction method based on multi-domain unified modeling according to claim 7 is characterized in that: In step 5, a fuzzy prediction model based on a four-layer neural network is built, which includes the following sub-steps: Step 5.1: Build the input layer, expressed as: in, represents the original input features, u ij is the data point x i The membership degree of cluster j, C represents the number of clusters; Step 5.2: Calculate the first layer. The calculation formula is: h1=σ(W1x i +b1) Where h1 represents the number of neurons in the first layer, b1 represents the bias term of the first layer, σ is the activation function, and W1 represents the weight matrix of the first layer; Step 5.3: Calculate the second and third layers using the following formula: Among them, W2 and W3 represent the weight matrices of the second and third layers, h2 and h3 represent the number of neurons in the second and third layers, and b2 and b3 represent the bias terms of the second and third layers; Step 5.4, build the output layer, the formula is: in, Represents the predicted value, W4 is the weight matrix of the output layer, and b4 is the bias term of the output layer; Step 5.4: Train the neural network and use MSE as the loss function, expressed as: in, is the predicted output of the neural network, y i is the actual output of the hybrid power generation system, and N is the number of data points.

9. An electronic device, characterized in that: include: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the hybrid power generation system energy prediction method based on multi-domain unified modeling as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the program is executed by a processor, the steps in the hybrid power generation system energy prediction method based on multi-domain unified modeling according to any one of claims 1 to 8 are implemented.

Images