Energy terminal multi-source data fusion analysis and self-adaptive regulation and control method and system

Abstract

The invention relates to the technical field of multi-source data fusion, in particular to an energy terminal multi-source data fusion analysis and self-adaptive regulation and control method and system. The method comprises the following steps: converting multi-source heterogeneous data into manifold coordinate data through complex manifold coordinate conversion and probability manifold measurement calculation; tensor fusion analysis is carried out on the flow form coordinate data based on a multi-core tensor kernel function, and low-dimensional fusion features with geometric structures reserved are generated; according to the invention, by collecting the operation state data of the equipment, comparing the deviation between the predicted value and the measured value of the fusion feature, and dynamically adjusting the multi-core tensor kernel function parameter of the fusion analysis module and the state equation parameter of the regulation and control decision module, a closed-loop correction mechanism is formed, and continuous self-optimization can be realized according to the actual operation condition of the equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-source data fusion, and particularly to a method and system for multi-source data fusion analysis and adaptive regulation of an energy terminal. Background Art

[0002] In the current era of rapid digital and intelligent development, energy management is crucial for achieving efficient energy utilization, reducing costs, and promoting sustainable development. As a key node for energy consumption and use, the operating status and data of the energy terminal contain rich information, and effective management and regulation of it are important research directions in the energy field.

[0003] Currently, for data processing and regulation of energy terminals, traditional multi-source data fusion analysis and adaptive regulation systems for energy terminals often adopt a single data source or a simple data fusion method. For example, only the power data of the power terminal is collected, and the start and stop of the device are controlled by simple threshold judgment; or when facing multi-source data, linear methods such as principal component analysis (PCA) are used for dimensionality reduction processing, and different types of energy terminal data (such as power and heat data) are simply weighted and fused. However, the multi-source heterogeneous data of energy terminals has a complex geometric structure, and there are non-linear relationships between the data. This traditional method cannot effectively process the complex geometric structure of multi-source heterogeneous data, and simple linear fusion will lose the non-linear relationships and important features between the data, resulting in insufficient utilization of data information and difficulty in accurately reflecting the true operating status of the energy terminal, thereby affecting the efficient management and regulation of the energy terminal. Summary of the Invention

[0004] To solve the above-mentioned problems, the present invention provides a method and system for multi-source data fusion analysis and adaptive regulation of an energy terminal. It solves the problems that traditional multi-source data fusion analysis and adaptive regulation systems for energy terminals have insufficient utilization of data information and difficulty in accurately reflecting the true operating status of the energy terminal, thereby affecting the efficient management and regulation of the energy terminal.

[0005] In the first aspect, a method for multi-source data fusion analysis and adaptive regulation of an energy terminal provided by the present invention adopts the following technical solution: A method for multi-source data fusion analysis and adaptive regulation of an energy terminal includes: Obtain multi-source heterogeneous data of the energy terminal; Through complex manifold coordinate transformation and probability manifold metric calculation, convert the multi-source heterogeneous data into manifold coordinate data; Based on a multi-core tensor kernel function, perform tensor fusion analysis on the manifold coordinate data to generate low-dimensional fusion features that retain the geometric structure; Parametrically model the device operation dynamics of the energy terminal using low-dimensional fusion features, and obtain the control strategy by solving the minimum condition of the stochastic Hamiltonian function; Generate device control instructions according to the control strategy; Perform adaptive optimization based on the feedback of the device operation state under the control instruction.

[0006] Furthermore, the acquisition of multi-source heterogeneous data of the energy terminal includes acquiring multi-source heterogeneous data of the energy terminal through a sensor array and the IEEE 1588 time synchronization protocol. The multi-source heterogeneous data includes complex voltage phasor data of the power terminal, temperature and humidity joint distribution data of the thermal terminal, and vibration signal data of the device. Based on complex manifold coordinate transformation and probability manifold metric calculation, the multi-source heterogeneous data is converted into manifold coordinate data.

[0007] Furthermore, the conversion of multi-source heterogeneous data into manifold coordinate data through complex manifold coordinate transformation and probability manifold metric calculation includes performing a holomorphic mapping transformation on the complex voltage phasor data. The transformation formula for performing the holomorphic mapping transformation is: , where, is the complex voltage phasor, is the real part, is the imaginary part; By converting the complex voltage phasor data into two independent real-valued components of the real part and the imaginary part, it is convenient for subsequent processing and realizes the formal conversion of the complex voltage phasor data.

[0008] Furthermore, the conversion of multi-source heterogeneous data into manifold coordinate data through complex manifold coordinate transformation and probability manifold metric calculation also includes, for the temperature and humidity joint distribution data, constructing a Fisher information matrix as the Riemannian metric. By using the Fisher information matrix as the metric tensor on the Riemannian manifold, the temperature and humidity joint distribution data is mapped onto the Riemannian manifold, so that the geometric structure of the data is preserved. The elements of the Fisher information matrix are expressed as: , where, is the distribution parameter, is the joint probability density function, .

[0009] Further, the tensor fusion analysis of the manifold coordinate data based on the multi-kernel tensor kernel function includes constructing the manifold coordinate data into a joint data tensor and calculating a kernel matrix based on the multi-kernel tensor kernel function to obtain a multi-kernel tensor kernel matrix. The analysis unit is used to perform centering processing and eigenvalue decomposition on the multi-kernel tensor kernel matrix to obtain eigenvectors and eigenvalues. The generation unit is used to perform embedding mapping calculation on the eigenvalues and eigenvectors based on the eigenvectors corresponding to the preset number of maximum eigenvalues to generate low-dimensional fusion features that retain the geometric structure. The multi-kernel tensor kernel function is defined as the product of the sub-manifold kernel functions, expressed as: , wherein, is the th joint data tensor, is the number of heterogeneous data types, is the th type of sub-manifold kernel function, is the th type of data.

[0010] Further, the tensor fusion analysis of the manifold coordinate data based on the multi-kernel tensor kernel function further includes performing centering processing and eigenvalue decomposition on the multi-kernel tensor kernel matrix to obtain eigenvectors and eigenvalues. Among them, the mean influence of the data is eliminated through centering processing, so that the data is analyzed subsequently under zero mean. The eigenvalue problem solved by eigenvalue decomposition is defined as , wherein, is the centered kernel matrix, is the centering matrix, is the eigenvector, is the eigenvalue, is the number of data samples. Through eigenvalue decomposition, the multi-kernel tensor kernel matrix is decomposed into the form of eigenvectors and eigenvalues. The eigenvalue reflects the variance size of the data in different directions. The larger the variance, the richer the data information in that direction.

[0011] Further, the parametric modeling of the device operation dynamics of the energy terminal using the low-dimensional fusion features includes parametrically modeling the device operation dynamics of the energy terminal by establishing a feature-parametric stochastic differential equation based on the mapping relationship between the low-dimensional fusion features and the device physical characteristics of the energy terminal. Among them, the low-dimensional fusion features retain the key geometric information of the multi-source heterogeneous data, and the device physical characteristics include the electrical parameters and mechanical characteristics of the device; by analyzing the mapping relationship, a state equation is established, wherein, is the state vector, is the cumulative energy consumption, is the device health state, is the cumulative carbon emissions, reflecting the impact of equipment operation on the environment; is the real-time power input, is a deterministic dynamic function; is a stochastic perturbation matrix; is a Wiener process vector.

[0012] Further, the control strategy obtained by solving the minimum condition of the stochastic Hamiltonian function includes defining a multi-objective performance functional containing energy consumption, equipment life, and carbon emissions based on the state equation, obtaining a performance functional expression containing expected integrals, constructing a stochastic Hamiltonian function based on the performance functional expression, and solving the minimum condition to obtain the control strategy. Among them, the performance functional expression is: , where, is the electricity price weight parameter; is the equipment maintenance cost weight parameter; is the carbon emission weight parameter; is the terminal health constraint weight parameter; is the real-time electricity price, is the regulation period.

[0013] Further, the generation of equipment control instructions according to the control strategy includes rolling optimization of the control strategy within the prediction time domain based on the model predictive control algorithm to obtain a reference power curve; designing a sliding mode surface and an exponential reaching law for the deviation between the actual operating state of the equipment at the energy terminal and the expected state corresponding to the reference power curve, and performing robust compensation for the unmodeled disturbance to obtain a robust compensation term. The sliding mode surface is defined as , where, is the state tracking error, is a positive definite matrix, and the exponential reaching law is , where, and are sliding mode control parameters; finally, linearly superimposing the reference power curve and the robust compensation term through the adaptive weight coefficient to generate equipment control instructions.

[0014] Further, the adaptive optimization based on the feedback of the equipment operating state under the control instruction includes collecting the operating state data of the terminal equipment in real time through sensors to obtain a real-time state data set, inputting the real-time state data set into a multi-core tensor kernel function, calculating the measured value of the fused feature, predicting the fused feature prediction value at the same time step based on the state equation, calculating the deviation between the two, and dynamically adjusting the parameters of the multi-core tensor kernel function and the parameters of the state equation based on the deviation to form a closed-loop correction mechanism. The calculation formula for the deviation is , where, is the fused feature prediction value, It is the measured value of the fusion feature.

[0015] In a second aspect, an energy terminal multi-source data fusion analysis and adaptive regulation system includes: A data acquisition module configured to acquire multi-source heterogeneous data of an energy terminal; A popular coordinate module configured to convert the multi-source heterogeneous data into manifold coordinate data through complex manifold coordinate transformation and probability manifold metric calculation; A feature fusion module configured to perform tensor fusion analysis on the manifold coordinate data based on a multi-core tensor kernel function to generate low-dimensional fusion features that retain the geometric structure; A control strategy module configured to parametrically model the device operation dynamics of the energy terminal using the low-dimensional fusion features, and obtain a control strategy by solving the minimum condition of the stochastic Hamiltonian function; A control instruction module configured to generate device control instructions according to the control strategy; A feedback module configured to perform adaptive optimization based on the feedback of the device operation state under the control instruction.

[0016] In a third aspect, the present invention provides a computer-readable storage medium storing multiple instructions, and the instructions are adapted to be loaded and executed by a processor of a terminal device for the method of multi-source data fusion analysis and adaptive regulation of an energy terminal.

[0017] In a fourth aspect, the present invention provides a terminal device including a processor and a computer-readable storage medium, where the processor is used to implement each instruction; the computer-readable storage medium is used to store multiple instructions, and the instructions are adapted to be loaded and executed by the processor for the method of multi-source data fusion analysis and adaptive regulation of an energy terminal.

[0018] In summary, the present invention has the following beneficial technical effects: 1. By acquiring multi-source heterogeneous data of the energy terminal, performing complex manifold coordinate transformation and probability manifold metric calculation to obtain manifold coordinate data, generating low-dimensional fusion features that retain the geometric structure based on a multi-core tensor kernel function, constructing a state equation and a multi-objective performance functional including a random perturbation term using the features, obtaining a control strategy by solving the stochastic Hamiltonian function, and simultaneously implementing the strategy, collecting the device operation state and adaptively optimizing the state equation parameters to form a closed-loop correction mechanism, the present invention realizes the deep fusion and accurate analysis of multi-source data, fully utilizes the data information and reflects the true operation state of the device, combines the control strategy and adaptive optimization considering multiple objectives and uncertainties, and solves the problems of insufficient utilization of data information in the traditional energy terminal multi-source data fusion analysis and adaptive regulation system, difficulty in accurately reflecting the true operation state of the energy terminal, and thus affecting the efficient management and regulation of the energy terminal.

[0019] 2. Through complex manifold coordinate transformation and probability manifold metric calculation, the present invention converts multi-source heterogeneous data into manifold coordinate data. For complex voltage phasor data, holomorphic mapping transformation is adopted, and for temperature and humidity joint distribution data, a Fisher information matrix is constructed as the Riemannian metric. Then, a multi-kernel tensor kernel function is used for tensor fusion analysis to more completely capture the complex non-linear relationships between data, enabling the low-dimensional fusion features to more completely reflect the operating state of the energy terminal, and enhancing the utilization value of the data.

[0020] 3. By centering and eigenvalue decomposition of the multi-kernel tensor kernel matrix, and based on the eigenvectors corresponding to the preset number of maximum eigenvalues, embedding mapping calculation is performed on the eigenvalues and eigenvectors to generate low-dimensional fusion features that retain the geometric structure. Thus, while retaining key information, the data dimension is reduced, the computational complexity is lowered, and the efficiency of data processing and analysis is improved, enabling the system to operate more quickly and efficiently when processing large-scale multi-source heterogeneous data.

[0021] 4. By constructing a state equation containing a random perturbation term and comprehensively considering factors such as energy consumption, equipment life, and carbon emissions to define a multi-objective performance functional, the formulation of the control strategy can take into account the multi-faceted requirements of energy consumption, equipment maintenance, environmental protection, etc., to maximize the comprehensive benefits of energy terminal equipment. The stochastic Hamiltonian function is used to solve the minimum condition to obtain the control strategy, which considers the random perturbations during the operation of the equipment, improves the robustness of the control strategy, and ensures the reliable operation of the energy terminal equipment.

[0022] 5. By combining model predictive control and sliding mode control, the present invention uses rolling optimization to obtain a reference power curve, designs a sliding mode surface and an exponential reaching law for robust compensation for the state deviation, and the two are weighted and superimposed to generate a control command. It can not only optimize using the forward-looking nature of model predictive control but also use sliding mode control to cope with unmodeled perturbations, effectively improving the accuracy and reliability of the control of energy terminal equipment and ensuring the stable and efficient operation of the equipment.

[0023] 6. By collecting the equipment operation state data, comparing the deviation between the predicted value and the measured value of the fusion features, and dynamically adjusting the parameters of the multi-kernel tensor kernel function of the fusion analysis module and the parameters of the state equation of the regulation decision module, a closed-loop correction mechanism is formed, which can continuously self-optimize according to the actual operation situation of the equipment, continuously improve the prediction and control capabilities of the operation state of the energy terminal equipment, enhance the long-term adaptability and regulation reliability, and further improve the performance and stability of the entire energy terminal regulation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of a method for multi-source data fusion analysis and adaptive regulation of an energy terminal according to Embodiment 1 of the present invention. Detailed implementation manners

[0025] The present invention will be further described in detail below with reference to the accompanying drawings.

[0026] Embodiment 1 Referring to Figure 1 , a multi-source data fusion analysis and adaptive regulation method for an energy terminal in this embodiment includes: Obtaining multi-source heterogeneous data of the energy terminal; Converting the multi-source heterogeneous data into manifold coordinate data through complex manifold coordinate transformation and probability manifold metric calculation; Performing tensor fusion analysis on the manifold coordinate data based on a multi-core tensor kernel function to generate low-dimensional fusion features that retain the geometric structure; Using the low-dimensional fusion features to perform parametric modeling on the device operation dynamics of the energy terminal, and obtaining a control strategy by solving the minimum condition of the stochastic Hamiltonian function; Generating device control instructions according to the control strategy; Performing adaptive optimization based on the feedback of the device operation state under the control instruction.

[0027] Specifically: S1. Obtaining multi-source heterogeneous data of the energy terminal, Among them, the acquisition and processing module includes a data acquisition unit and a data processing unit. The data acquisition unit is used to obtain multi-source heterogeneous data of the energy terminal through a sensor array and the IEEE 1588 time synchronization protocol. The multi-source heterogeneous data includes complex voltage phasor data of the power terminal, temperature and humidity joint distribution data of the thermal terminal, and vibration signal data of the device. The data processing unit is used to convert the multi-source heterogeneous data into manifold coordinate data based on complex manifold coordinate transformation and probability manifold metric calculation.

[0028] S2. Converting the multi-source heterogeneous data into manifold coordinate data through complex manifold coordinate transformation and probability manifold metric calculation; Among them, when the data is converted into manifold coordinate data, a holomorphic mapping transformation is performed on the complex voltage phasor data, and a Fisher information matrix is constructed as the Riemannian metric for the temperature and humidity joint distribution data; The transformation formula for performing the holomorphic mapping transformation is: , where is the complex voltage phasor, is the real part, is the imaginary part; The elements of the Fisher information matrix are: , where is a distribution parameter, is the joint probability density function, .

[0029] Specifically, through complex manifold coordinate transformation and probability manifold metric calculation, multi-source heterogeneous data is converted into manifold coordinate data. This module includes a data acquisition unit and a data processing unit. The data acquisition unit obtains multi-source heterogeneous data of the energy terminal through a sensor array and the IEEE 1588 time synchronization protocol. These data include complex voltage phasor data of the power terminal, temperature-humidity joint distribution data of the thermal terminal, and vibration signal data of the equipment. The complex voltage phasor data of the power terminal is collected through a voltage transformer and a current transformer, the temperature-humidity joint distribution data of the thermal terminal is obtained by a temperature-humidity sensor network, and the vibration signal data of the equipment is collected through an acceleration sensor. The raw data collected by the sensor array is time-synchronized through the IEEE 1588 time synchronization protocol to ensure the consistency of multi-source data in the time dimension.

[0030] Among them, (1) The data processing unit converts multi-source heterogeneous data into manifold coordinate data based on complex manifold coordinate transformation and probability manifold metric calculation. For complex voltage phasor data, a holomorphic mapping transformation is adopted. A holomorphic mapping is an important concept in complex analysis. It preserves the complex structure and can effectively process complex numbers in the power system. The transformation formula for performing the holomorphic mapping transformation is: , where, is the complex voltage phasor, is the real part, is the imaginary part. Through this formula, the complex voltage phasor data is converted into two independent real-valued components, the real part and the imaginary part, which is convenient for subsequent processing. In the processing of complex voltage phasor data, assume the input complex voltage phasor , according to the holomorphic mapping transformation formula , where is the real part, that is, 3; is the imaginary part, that is, 4. The output is , which converts the original complex voltage phasor data into two independent real-valued components, the real part and the imaginary part, facilitating subsequent operations such as analysis and processing of power terminal data, realizing the formal transformation of complex voltage phasor data, and making it more suitable for subsequent data processing flows.

[0031] (2) For temperature-humidity joint distribution data, a Fisher information matrix is constructed as the Riemannian metric. The Fisher information matrix is an important tool in statistics for measuring the accuracy of parameter estimation. In manifold learning, it can be used as the metric tensor on the Riemannian manifold. The elements of the Fisher information matrix are: , Among them, is a distribution parameter, is the joint probability density function, .

[0032] By constructing the Fisher information matrix, the joint temperature and humidity distribution data is mapped onto a Riemannian manifold, enabling the preservation of the geometric structure of the data. In the processing of joint temperature and humidity distribution data, let the distribution parameter , where is the temperature, is the humidity, and the joint probability density function is known. Assume that the joint probability density function of temperature and humidity within a certain range is (only an example function). For the elements of the Fisher information matrix: , When , first find , Take the logarithm of to get , Then , and similarly .

[0033] Substitute into the formula for calculation: , After performing the integration operation (assuming the integration range is ), the value of can be obtained.

[0034] By calculating the values of each element of the Fisher information matrix, the Fisher information matrix is constructed, mapping the joint temperature and humidity distribution data onto a Riemannian manifold, preserving the geometric structure of the joint temperature and humidity distribution data, and providing a more suitable data form for subsequent analysis and processing of thermal terminal data.

[0035] (3) For the vibration signal data of the device, first perform preprocessing, including operations such as denoising and filtering, and then extract characteristic parameters such as vibration frequency and amplitude. These characteristic parameters constitute the characteristic space of the vibration signal, and through the manifold learning algorithm, it is mapped into a low-dimensional manifold space to achieve data dimensionality reduction and feature extraction.

[0036] Through the above complex manifold coordinate transformation and probability manifold metric calculation, multi-source heterogeneous data is transformed into unified manifold coordinate data. This transformation method can effectively preserve the internal geometric structure of the data and provide a good data foundation for subsequent fusion analysis. In the actual application of the energy terminal, this acquisition and processing module can improve the accuracy and efficiency of data processing and provide strong support for the optimal regulation of the energy terminal.

[0037] S3. Perform tensor fusion analysis on the manifold coordinate data based on the multi-core tensor kernel function to generate low-dimensional fusion features that retain the geometric structure. Among them, based on the multi-core tensor kernel function, the manifold coordinate data is constructed into a joint data tensor and the kernel matrix is calculated to obtain the multi-core tensor kernel matrix. The analysis unit is used to perform centering processing and eigenvalue decomposition on the multi-core tensor kernel matrix to obtain eigenvectors and eigenvalues. The generation unit is used to perform embedding mapping calculation on the eigenvalues and eigenvectors based on the eigenvectors corresponding to the preset number of the largest eigenvalues to generate low-dimensional fusion features that retain the geometric structure.

[0038] The multi-core tensor kernel function is defined as the product of the sub-manifold kernel functions: , where is the th joint data tensor, is the number of heterogeneous data types, is the th sub-manifold kernel function of the th th type of data; The eigenvalue problem solved by eigenvalue decomposition is , where is the centered kernel matrix, is the centering matrix, is the eigenvector, is the eigenvalue, is the number of data samples.

[0039] Specifically, (1) The fusion analysis module is used to perform tensor fusion analysis on the manifold coordinate data based on the multi-core tensor kernel function to generate low-dimensional fusion features that retain the geometric structure. This module includes a fusion unit, an analysis unit, and a generation unit.

[0040] The fusion unit constructs the manifold coordinate data into a joint data tensor and calculates the kernel matrix based on the multi-core tensor kernel function to obtain the multi-core tensor kernel matrix. The multi-core tensor kernel function is defined as the product of the sub-manifold kernel functions, that is: , where is the a combined data tensor, is the number of heterogeneous data types, is the class of submanifold kernel functions, is the th class of data. Different types of submanifold kernel functions are selected according to the characteristics of the data, such as Gaussian kernel functions, polynomial kernel functions, etc. Taking three types of heterogeneous data, i.e., the complex voltage phasor data of power terminals, the joint temperature and humidity distribution data of thermal terminals, and the vibration signal data of equipment, as examples, they respectively correspond to different submanifold kernel functions. The coordinate data of these three types of manifolds are constructed into a combined data tensor, and the kernel matrix is calculated through the multi-core tensor kernel function to capture the complex non-linear relationships between different types of data.

[0041] Among them, for the multi-core tensor kernel function formula: , Let the number of heterogeneous data types , which are respectively the complex voltage phasor data of power terminals (corresponding to ), the joint temperature and humidity distribution data of thermal terminals (corresponding to ), and the vibration signal data of equipment (corresponding to ). Assume are two combined data tensors, where is the complex voltage phasor data, is the joint temperature and humidity distribution data, is the vibration signal data. Let the first-class submanifold kernel function be the Gaussian kernel function , the second-class submanifold kernel function be the polynomial kernel function , and the third-class submanifold kernel function be the linear kernel function . Substituting the above data into the multi-core tensor kernel function formula, we can get . After calculation, an element value in the kernel matrix is obtained. By calculating different combined data tensors, a multi-core tensor kernel matrix is constructed. This process realizes the fusion calculation of coordinate data of different types of manifolds, captures the complex non-linear relationships between different types of data, and provides a basis for subsequent analysis.

[0042] (2) The analysis unit performs centering processing and eigenvalue decomposition on the multi-core tensor kernel matrix to obtain eigenvectors and eigenvalues. Centering processing is to eliminate the influence of the data mean and enable subsequent analysis of the data under zero mean. The eigenvalue problem solved by eigenvalue decomposition is , where is the centered kernel matrix, is a centralized matrix, is an eigenvector, is an eigenvalue, is the number of data samples. Through eigenvalue decomposition, the multi-core tensor kernel matrix is decomposed into the form of eigenvectors and eigenvalues. The eigenvalues reflect the variance of the data in different directions. The larger the variance, the richer the data information in that direction.

[0043] Specifically, for the eigenvalue problem formula , let the number of data samples , the centralized kernel matrix is a matrix, the centralized matrix , where is the identity matrix, is the all-ones column vector. Assume that the multi-core tensor kernel matrix has been obtained through the fusion unit. Perform eigenvalue decomposition on this matrix to solve the equation through an iterative algorithm (such as the power method, etc.), and obtain the eigenvector and the eigenvalue . For example, obtain eigenvalues etc. and their corresponding eigenvectors etc. The eigenvalues reflect the variance of the data in different directions. Through this eigenvalue decomposition, the multi-core tensor kernel matrix is decomposed into the form of eigenvectors and eigenvalues, extracting the key information of the data in different dimensions, providing a basis for generating low-dimensional fusion features later, and realizing the extraction of data features of the multi-core tensor kernel matrix.

[0044] (3) Based on the eigenvectors corresponding to the preset number of the largest eigenvalues, perform embedding mapping calculations on the eigenvalues and eigenvectors to generate low-dimensional fusion features that retain the geometric structure. Because the directions of the eigenvectors corresponding to the largest eigenvalues contain the main information of the data, by performing embedding mapping calculations on these eigenvalues and eigenvectors, the high-dimensional manifold coordinate data is mapped to a low-dimensional space while retaining the geometric structure of the data. Such low-dimensional fusion features not only reduce the data dimension and computational complexity but also retain the key geometric information of the original multi-source heterogeneous data, providing a more effective data representation for subsequent regulation and decision-making. Through the processing of the fusion analysis module, the multi-source heterogeneous data of the energy terminal can be presented in a more compact and representative form, improving the utilization efficiency of the data and laying a good foundation for the optimal regulation of the energy terminal.

[0045] S4. Parametrically model the device operation dynamics of the energy terminal using low-dimensional fusion features, construct a state equation including a stochastic perturbation term, and define a multi-objective performance functional including energy consumption, device lifespan, and carbon emissions. Obtain the control strategy by solving the minimum condition of the stochastic Hamiltonian function. The regulation decision-making module includes a stochastic state modeling unit, a performance functional definition unit, and a stochastic Hamiltonian solving unit. The stochastic state modeling unit is used to parametrically model the device operation dynamics of the energy terminal based on the mapping relationship between low-dimensional fusion features and the physical characteristics of the devices in the energy terminal. By establishing a parametric stochastic differential equation, a state equation including a stochastic perturbation term is constructed. The performance functional definition unit is used to define a multi-objective performance functional including energy consumption, device lifespan, and carbon emissions based on the state equation, and obtain a performance functional expression including expected integrals. The stochastic Hamiltonian solving unit is used to construct a stochastic Hamiltonian function based on the performance functional expression and solve the minimum condition to obtain the control strategy.

[0046] (1) The state equation is: , where is the state vector, is the cumulative energy consumption, is the device health state, is the cumulative carbon emissions, is the real-time power input, is the deterministic dynamic function, is the stochastic perturbation matrix, is the Wiener process vector; The performance functional expression is: , where is the electricity price weight parameter, is the device maintenance cost weight parameter, is the carbon emissions weight parameter, is the terminal health constraint weight parameter, is the real-time electricity price, is the regulation period.

[0047] Specifically, the regulation decision-making module parametrically models the device operation dynamics of the energy terminal using low-dimensional fusion features, constructs a state equation including a stochastic perturbation term, and defines a multi-objective performance functional including energy consumption, device lifespan, and carbon emissions. The control strategy is obtained by solving the minimum condition of the stochastic Hamiltonian function. This module includes a stochastic state modeling unit, a performance functional definition unit, and a stochastic Hamiltonian solving unit.

[0048] The random state modeling unit parametrically models the operation dynamics of the energy terminal devices by establishing a parametric stochastic differential equation based on the mapping relationship between the low-dimensional fusion features and the device physical characteristics of the energy terminal. The low-dimensional fusion features are obtained by processing through the fusion analysis module, which retains the key geometric information of the multi-source heterogeneous data. The device physical characteristics include the electrical parameters, mechanical characteristics, etc. of the device. By analyzing the mapping relationship between the two, a state equation is established. . Among them, is the state vector, is the cumulative energy consumption, reflecting the energy consumption during the device operation; is the device health state, measuring the aging, wear, etc. of the device; is the cumulative carbon emission, reflecting the impact of the device operation on the environment. is the real-time power input, is the deterministic dynamic function, describing the operation law of the device under deterministic factors; is the random perturbation matrix, considering uncertain factors such as environmental factors and accidental device failures; is the Wiener process vector, representing random noise. This state equation comprehensively describes the operation dynamics of the energy terminal devices and provides a basic model for subsequent decision-making.

[0049] For the state equation formula , assume the input state vector , where is the cumulative energy consumption, is the device health state, is the cumulative carbon emission; the control input is the real-time power; the deterministic dynamic function , where is the device degradation coefficient, is the carbon emission coefficient; the random perturbation matrix , where are the random perturbation coefficients of energy consumption, device health state, and carbon emission respectively; is the Wiener process vector. Through this state equation, combined with the physical characteristics and operation environment of the device, parametric modeling of the device operation dynamics is carried out, and the state prediction of the device in the future period is output, including energy consumption change, device health state evolution, and carbon emission growth, etc., obtaining a dynamic model of the device operation, realizing the accurate description and prediction of the operation state of the energy terminal devices, and providing a basis for subsequent control decisions.

[0050] (2) The performance functional definition unit defines a multi-objective performance functional including energy consumption, device life, and carbon emission based on the state equation. The performance functional expression is: , Among them, is the electricity price weight parameter, which reflects the importance of the electricity price in the objective function; is the equipment maintenance cost weight parameter, which reflects the impact of the equipment maintenance cost; is the carbon emission weight parameter, which measures the cost of carbon emissions; is the terminal health constraint weight parameter, which emphasizes the importance of the equipment health status; is the real-time electricity price, is the regulation period. This performance functional comprehensively considers the economic costs (electricity price, equipment maintenance cost) and environmental impacts (carbon emissions) as well as the equipment's own health status, and comprehensively measures the comprehensive benefits of the operation of energy terminal equipment.

[0051] For the performance functional expression: , Let the input parameter be the electricity price weight parameter, be the equipment maintenance cost weight parameter, be the carbon emission weight parameter, be the terminal health constraint weight parameter, be the power, be the real-time electricity price, be the equipment health status, be the carbon emission, be the regulation period. Through this performance functional expression, considering multiple factors such as energy consumption cost, equipment life loss, and carbon emissions, a comprehensive performance evaluation value is output. This value reflects the comprehensive benefits of the equipment operation during the entire regulation period, obtaining an index that can comprehensively measure the equipment operation effect, achieving multi-objective optimization evaluation of the operation of energy terminal equipment, and providing a basis for finding the optimal control strategy.

[0052] (3) The stochastic Hamiltonian solution unit constructs a stochastic Hamiltonian function and solves the minimum condition based on the performance functional expression to obtain the control strategy. The stochastic Hamiltonian function is an important tool for solving stochastic optimal control problems. By solving its minimum condition, the control strategy that optimizes the performance functional considering stochastic factors can be obtained. This process involves complex mathematical derivations and solutions. Using stochastic analysis theory and optimal control algorithms, the optimal values of control variables such as the real-time power input are determined, thereby realizing the optimal regulation of energy terminal equipment. Through the processing of the regulation decision module, a reasonable control strategy can be formulated considering multiple factors and stochastic perturbations, realizing the efficient, economic, and environmentally friendly operation of energy terminal equipment.

[0053] S5. Control Execution Module: It is used to generate and execute device control instructions according to the control strategy; the control execution module includes a predictive control unit, a sliding mode control unit, a generation unit, and an execution unit. The predictive control unit is used to perform rolling optimization of the control strategy within the prediction time domain based on the model predictive control algorithm to obtain a reference power curve. The sliding mode control unit is used to design a sliding mode surface and an exponential reaching law for the deviation between the actual operating state of the device at the energy terminal and the desired state corresponding to the reference power curve, and perform robust compensation for the unmodeled disturbances to obtain a robust compensation term. The sliding mode surface is defined as , where is the state tracking error, is a positive definite matrix, and the exponential reaching law is , where and are the sliding mode control parameters, and the unmodeled disturbances are the device nonlinear dynamics and external environment disturbances not included in the model predictive control algorithm. The generation unit is used to linearly superpose the reference power curve and the robust compensation term through the adaptive weight coefficient to generate a device control instruction. The execution unit is used to convert the device control instruction into an analog signal through a digital-to-analog conversion module and transmit it to the device at the energy terminal, so that the device at the energy terminal executes the device control instruction.

[0054] Specifically, the control execution module is used to generate and execute device control instructions according to the control strategy, and it is composed of a predictive control unit, a sliding mode control unit, a generation unit, and an execution unit.

[0055] (1) The predictive control unit performs rolling optimization of the control strategy within the prediction time domain based on the model predictive control algorithm. The model predictive control algorithm predicts the operating state of the device in the future for a period of time according to the dynamic model of the device at the energy terminal. In each control cycle, this unit will optimize the control strategy according to the actual operating state of the current device, combined with the prediction model. For example, considering factors such as the real-time power demand of the device at the energy terminal, the device health status, and external environment changes, the control strategy is continuously adjusted within the prediction time domain to obtain the optimal reference power curve. This rolling optimization method can timely respond to various changes during the device operation, improving the accuracy and adaptability of the control.

[0056] The sliding mode control unit works for the deviation between the actual operating state of the device at the energy terminal and the desired state corresponding to the reference power curve. During the actual operation of the device, due to the existence of unmodeled disturbances, such as device nonlinear dynamics and external environment disturbances, the actual state will deviate from the desired state. The sliding mode control unit solves this problem by designing a sliding mode surface and an exponential reaching law. The sliding mode surface is defined as , where is the state tracking error, that is, the difference between the actual state and the desired state, is a positive definite matrix, which determines the speed and characteristics of the system state converging to the sliding mode surface. The exponential reaching law is , where and are sliding mode control parameters. By reasonably selecting these parameters, the system state can approach the sliding mode surface quickly and stably, perform robust compensation for unmodeled disturbances, obtain a robust compensation term, thereby enhancing the system's resistance to uncertain factors.

[0057] The generation unit linearly superposes the reference power curve and the robust compensation term through the adaptive weight coefficient . The adaptive weight coefficient will be dynamically adjusted according to factors such as the real-time operating state of the device and changes in the external environment. For example, when the device is operating relatively stably, the weight of the reference power curve in the control instruction will be relatively large; when the device is subject to large disturbances and the deviation between the actual state and the desired state is large, the weight of the robust compensation term will increase accordingly, so as to generate a device control instruction to ensure that the control instruction can both follow the expected optimization strategy and effectively respond to sudden disturbance situations.

[0058] (2) The execution unit converts the device control instruction into an analog signal through the digital-to-analog conversion module and transmits it to the device of the energy terminal. The digital-to-analog conversion module converts the digital control instruction into an analog signal suitable for the device to receive, such as a voltage or current signal, so that the device of the energy terminal can execute the device control instruction, achieve precise adjustment of the device operating state, and ensure the stable and efficient operation of the energy terminal device.

[0059] S6. State feedback module: It is used to collect the operating state of the energy terminal device and feedback it to the regulation and decision-making module to adaptively optimize the parameters of the state equation. The state feedback module includes a state acquisition unit and a feedback optimization unit. The state acquisition unit is used to collect the operating state data of the terminal device in real time through sensors to obtain a real-time state data set. The sensors include current sensors, temperature sensors, and vibration sensors. The operating state data includes power, temperature, and health index. The feedback optimization unit is used to input the real-time state data set into the multi-core tensor kernel function of the fusion analysis module, calculate the measured value of the fusion feature, predict the predicted value of the fusion feature at the same time step based on the state equation of the regulation and decision-making module, calculate the deviation between the two, and based on the deviation, dynamically adjust the parameters of the multi-core tensor kernel function in the fusion analysis module and the parameters of the state equation in the regulation and decision-making module to form a closed-loop correction mechanism. The calculation formula for the deviation is , where is the predicted value of the fusion feature, is the measured value of the fusion feature.

[0060] Specifically, the state feedback module is used to collect the operating state of the energy terminal device and feedback it to the regulation decision-making module, and adaptively optimize the parameters of the state equation. It consists of a state acquisition unit and a feedback optimization unit.

[0061] The state acquisition unit collects the operating state data of the terminal device in real time through current sensors, temperature sensors, vibration sensors, etc., to obtain a real-time state data set. These sensors are distributed at key parts of the energy terminal device. The current sensor is used to monitor the current change of the device to obtain power information; the temperature sensor perceives the temperature of the device in real time to reflect the operating load and health status of the device; the vibration sensor captures the vibration signal of the device, which can be used to judge whether there are mechanical failures or other abnormalities in the device. The collected operating state data includes power, temperature, health index, etc. These data constitute the real-time state data set, providing a basis for subsequent analysis.

[0062] The feedback optimization unit inputs the real-time state data set into the multi-core tensor kernel function of the fusion analysis module. The multi-core tensor kernel function is a key tool in the fusion analysis module for processing multi-source heterogeneous data, which can fuse and analyze different types of state data. Through the calculation of the multi-core tensor kernel function, the measured value of the fusion feature is obtained. At the same time, based on the state equation of the regulation decision-making module, using the currently known parameters and models, the predicted value of the fusion feature at the same time step is predicted. Here, the state equation is a mathematical model used by the regulation decision-making module to describe the operation dynamics of the device, which contains key information such as the energy consumption, health status, and carbon emissions of the device.

[0063] Calculate the deviation between the predicted value and the measured value of the fusion feature. The formula is , where is the predicted value of the fusion feature, is the measured value of the fusion feature. Based on this deviation, dynamically adjust the parameters of the multi-core tensor kernel function in the fusion analysis module and the parameters of the state equation in the regulation decision-making module. For example, if the deviation is large, it means that there is a large difference between the current model parameters and the actual situation. At this time, through specific algorithms, such as the gradient descent method, etc., adjust the parameters of the multi-core tensor kernel function (such as the bandwidth of the kernel function, etc.) and the parameters of the state equation (such as the variance of the random perturbation matrix, etc.), so that the model can better fit the actual data, forming a closed-loop correction mechanism. This closed-loop correction mechanism can continuously optimize the system model, improve the prediction and control accuracy of the operating state of the energy terminal device, and make the entire energy terminal regulation system operate more stably and efficiently.

[0064] The multi-source heterogeneous data of the energy terminal is acquired by the acquisition and processing module, and through complex manifold coordinate transformation and probability manifold metric calculation, these data are converted into manifold coordinate data, laying a foundation for subsequent processing. The fusion analysis module performs tensor fusion analysis on the manifold coordinate data based on the multi-core tensor kernel function to generate low-dimensional fusion features that retain the geometric structure, effectively capturing the non-linear relationships between the data and avoiding the problem of data feature loss caused by traditional linear fusion. The regulation and decision-making module uses the low-dimensional fusion features to parametrically model the device operation dynamics of the energy terminal, constructs a state equation containing a random perturbation term, and defines a multi-objective performance functional covering energy consumption, device life, and carbon emissions. By solving the minimum condition of the stochastic Hamiltonian function, a control strategy is obtained, so as to formulate a reasonable control strategy considering various influencing factors and uncertainties. The control execution module generates and executes device control instructions according to the control strategy to ensure the effective implementation of the strategy. The state feedback module acquires the operation state of the energy terminal device and feeds it back to the regulation and decision-making module to adaptively optimize the parameters of the state equation, forming a closed-loop correction mechanism, enabling the system to continuously adjust and optimize according to the actual operation situation, thus realizing the deep processing and fusion of the multi-source heterogeneous data of the energy terminal, making full use of the effective information in the data, and accurately reflecting the true operation state of the energy terminal. On this basis, combined with the regulation and decision-making considering multi-objectives and uncertainties and the adaptive optimization mechanism, the efficient adaptive regulation of the energy terminal is realized. It effectively solves the problems in the traditional multi-source data fusion analysis and adaptive regulation system of the energy terminal, such as insufficient utilization of data information, difficulty in accurately reflecting the true operation state of the energy terminal, and further affecting the efficient management and regulation of the energy terminal, improves the energy utilization efficiency, extends the service life of the device, and provides strong support for the intelligent management of the energy terminal.

[0065] Embodiment 2 This embodiment provides a multi-source data fusion analysis and adaptive regulation system for an energy terminal, including: A data acquisition module configured to acquire multi-source heterogeneous data of the energy terminal; A manifold coordinate module configured to convert the multi-source heterogeneous data into manifold coordinate data through complex manifold coordinate transformation and probability manifold metric calculation; A feature fusion module configured to perform tensor fusion analysis on the manifold coordinate data based on the multi-core tensor kernel function to generate low-dimensional fusion features that retain the geometric structure; A control strategy module configured to parametrically model the device operation dynamics of the energy terminal using the low-dimensional fusion features and obtain a control strategy by solving the minimum condition of the stochastic Hamiltonian function; A control instruction module configured to generate device control instructions according to the control strategy; A feedback module, configured to perform adaptive optimization based on the feedback of the device operating state under a control instruction.

[0066] A computer-readable storage medium storing multiple instructions adapted to be loaded and executed by a processor of a terminal device for the method of multi-source data fusion analysis and adaptive regulation of an energy terminal.

[0067] A terminal device, comprising a processor and a computer-readable storage medium, the processor for implementing each instruction; the computer-readable storage medium for storing multiple instructions adapted to be loaded and executed by the processor for the method of multi-source data fusion analysis and adaptive regulation of an energy terminal.

[0068] The above are all preferred embodiments of the present invention. Without limiting the protection scope of the present invention accordingly, therefore: All equivalent changes made according to the structure, shape, and principle of the present invention shall be covered within the protection scope of the present invention.

Claims

1. A multi-source data fusion analysis and adaptive regulation method for an energy terminal, characterized in that, Including: Obtain multi-source heterogeneous data of the energy terminal; Through complex manifold coordinate transformation and probability manifold metric calculation, convert the multi-source heterogeneous data into manifold coordinate data; Based on the multi-core tensor kernel function, perform tensor fusion analysis on the manifold coordinate data to generate low-dimensional fusion features that retain the geometric structure; Use the low-dimensional fusion features to parametrically model the device operation dynamics of the energy terminal, and obtain the control strategy by solving the minimum condition of the stochastic Hamiltonian function; Generate device control instructions according to the control strategy; Perform adaptive optimization based on the feedback of the device operation state under the control instruction.

2. The multi-source data fusion analysis and adaptive regulation method for an energy terminal according to claim 1, characterized in that The conversion of multi-source heterogeneous data into manifold coordinate data through complex manifold coordinate transformation and probability manifold metric calculation includes performing a holomorphic mapping transformation on the complex voltage phasor data. The transformation formula for performing the holomorphic mapping transformation is: ， Among them, is the complex voltage phasor, is the real part, is the imaginary part; by converting the complex voltage phasor data into two independent real-valued components of the real part and the imaginary part, it is convenient for subsequent processing and realizes the formal conversion of the complex voltage phasor data.

3. A multi-source data fusion analysis and adaptive regulation method for an energy terminal according to claim 2, characterized in that The conversion of multi-source heterogeneous data into manifold coordinate data through complex manifold coordinate transformation and probability manifold metric calculation also includes constructing a Fisher information matrix as the Riemannian metric for the temperature and humidity joint distribution data. Using the Fisher information matrix as the metric tensor on the Riemannian manifold, map the temperature and humidity joint distribution data onto the Riemannian manifold so that the geometric structure of the data is retained. The elements of the Fisher information matrix are expressed as: ， wherein, is a distribution parameter, is a joint probability density function, .

4. A multi-source data fusion analysis and adaptive regulation method for an energy terminal according to claim 3, characterized in that The tensor fusion analysis of the manifold coordinate data based on the multi-core tensor kernel function includes, based on the multi-core tensor kernel function, constructing the manifold coordinate data into a joint data tensor and calculating the kernel matrix to obtain the multi-core tensor kernel matrix. The analysis unit is used to perform centering processing and eigenvalue decomposition on the multi-core tensor kernel matrix to obtain eigenvectors and eigenvalues. The generation unit is used to perform embedding mapping calculation on the eigenvalues and eigenvectors based on the eigenvectors corresponding to the preset number of the largest eigenvalues to generate low-dimensional fusion features that retain the geometric structure. The multi-core tensor kernel function is defined as the product of the sub-manifold kernel functions and is expressed as: ， Among them, is the th combined data tensor, is the number of heterogeneous data types, is the th class of submanifold kernel function, is the th class of data.

5. The multi-source data fusion analysis and adaptive regulation method for an energy terminal according to claim 4, characterized in that The tensor fusion analysis of the manifold coordinate data based on the multi-core tensor kernel function further includes centering processing and eigenvalue decomposition of the multi-core tensor kernel matrix to obtain eigenvectors and eigenvalues. Among them, the influence of the data mean is eliminated through centering processing, so that the subsequent analysis is carried out under zero mean. The eigenvalue problem solved by eigenvalue decomposition is defined as , where is the centered kernel matrix, is the centering matrix, is the eigenvector, is the eigenvalue, is the number of data samples. Through eigenvalue decomposition, the multi-core tensor kernel matrix is decomposed into the form of eigenvectors and eigenvalues. The eigenvalue reflects the variance size of the data in different directions. The larger the variance, the richer the data information in that direction.

6. The multi-source data fusion analysis and adaptive regulation method for an energy terminal according to claim 5, wherein The parametric modeling of the device operation dynamics of the energy terminal using low-dimensional fusion features includes parametrically modeling the device operation dynamics of the energy terminal by establishing a feature-parametric stochastic differential equation based on the mapping relationship between the low-dimensional fusion features and the device physical characteristics of the energy terminal. Among them, the low-dimensional fusion features retain the key geometric information of multi-source heterogeneous data, and the device physical characteristics include the electrical parameters and mechanical characteristics of the device; by analyzing the mapping relationship, a state equation is established , where is the state vector, is the cumulative energy consumption, is the device health state, is the cumulative carbon emission, reflecting the impact of device operation on the environment; is the real-time power input, is the deterministic dynamic function; is the stochastic perturbation matrix; is the Wiener process vector.

7. The multi-source data fusion analysis and adaptive regulation method for an energy terminal according to claim 6, characterized in that, The obtaining of the control strategy by solving the minimum condition of the stochastic Hamiltonian function includes, based on the state equation, defining a multi-objective performance functional including energy consumption, device life, and carbon emissions to obtain a performance functional expression including the expected integral. Based on the performance functional expression, construct a stochastic Hamiltonian function and solve the minimum condition to obtain the control strategy. The performance functional expression is: ， Among them, is the electricity price weight parameter; is the equipment maintenance cost weight parameter; is the carbon emission weight parameter; is the terminal health constraint weight parameter; is the real-time electricity price, is the regulation period.

8. A method for multi-source data fusion analysis and adaptive regulation of an energy terminal according to claim 7, characterized in that Generating the device control instruction according to the control strategy includes performing rolling optimization on the control strategy within the prediction time domain based on the model predictive control algorithm to obtain a reference power curve; designing a sliding surface and an exponential reaching law for the deviation between the actual operating state of the device at the energy terminal and the desired state corresponding to the reference power curve to perform robust compensation for the unmodeled disturbance and obtain a robust compensation term, where the sliding surface is defined as , where is the state tracking error, is a positive definite matrix, and the exponential reaching law is , where and are the sliding mode control parameters; finally, linearly superimposing the reference power curve and the robust compensation term through the adaptive weight coefficient to generate the device control instruction.

9. The multi-source data fusion analysis and adaptive regulation method for an energy terminal according to claim 8, characterized in that The adaptive optimization based on the feedback of the device operation state under the control instruction includes real-time collecting the operation state data of the terminal device through sensors to obtain a real-time state data set, inputting the real-time state data set into the multi-core tensor kernel function, calculating the measured value of the fusion feature, predicting the predicted value of the fusion feature at the same time step based on the state equation, calculating the deviation between the two, and dynamically adjusting the parameters of the multi-core tensor kernel function and the parameters of the state equation based on the deviation to form a closed-loop correction mechanism. The calculation formula for the deviation is: ， Among them, is the predicted value of the fusion feature, is the measured value of the fusion feature.

10. An energy terminal multi-source data fusion analysis and adaptive regulation system, characterized in that, Including: A data acquisition module configured to obtain multi-source heterogeneous data of the energy terminal; A manifold coordinate module configured to convert the multi-source heterogeneous data into manifold coordinate data through complex manifold coordinate transformation and probability manifold metric calculation; A feature fusion module, configured to perform tensor fusion analysis on manifold coordinate data based on a multi-core tensor kernel function to generate low-dimensional fusion features that preserve the geometric structure; A control strategy module, configured to parametrically model the device operation dynamics of an energy terminal using the low-dimensional fusion features, and obtain a control strategy by solving the minimum condition of a stochastic Hamiltonian function; A control instruction module, configured to generate device control instructions according to the control strategy; A feedback module, configured to perform adaptive optimization based on the feedback of the device operation state under the control instructions.

Images