Stability analysis and early warning method for building hydrogen-electricity coupling direct current micro-grid

Abstract

The invention belongs to the technical field of micro-grid stability analysis, and particularly relates to a stability analysis and early warning method for a building hydrogen-electricity coupling direct current micro-grid, and the method comprises the steps: obtaining an electromagnetic transient state quantity of an electrical subsystem in the direct current micro-grid and an electrochemical reaction state quantity of a hydrogen energy subsystem, and constructing a unified characteristic matrix of time scale dynamic characteristics; extracting a stability feature, projecting the stability feature to a low-dimensional space through nonlinear mapping, forming a representation fingerprint, and calculating an instantaneous stability tolerance in a current state; time sequence deduction is carried out, a stability evolution trajectory in a future time window is solved, and a prediction offset vector of an offset direction and an offset amplitude is generated; and when the predicted offset vector exceeds the instantaneous stability tolerance, taking minimization of the predicted offset vector as an optimization target, and generating a cooperative control strategy for cooperatively regulating and controlling the operating point of the hydrogen energy subsystem and the operating point of the electrical subsystem. According to the invention, sensing and early warning of the instability risk can be realized, and stable and reliable operation of the microgrid is guaranteed.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of microgrid stability analysis, and particularly relates to a stability analysis and early warning method for a building hydrogen-electricity coupled direct-current microgrid. BACKGROUND

[0002] The building hydrogen-electricity coupled direct-current microgrid is a new type of distributed energy system, which takes direct current as the main transmission form, integrates photovoltaic renewable energy, energy storage devices and various types of direct current loads in the building. In order to solve the intermittency of renewable energy and the long-time energy storage demand, the hydrogen energy subsystem is introduced as a long-time energy storage technology, which forms a complex energy interaction network with the electrical subsystem through the hydrogen storage by electrolysis and the power generation by fuel cells.

[0003] In the prior art, for the stability analysis of such a microgrid, the electrical subsystem and the hydrogen energy subsystem are usually modeled separately. For the electrical subsystem with fast response speed, electromagnetic transient simulation is mainly used to analyze the dynamic characteristics of voltage and current; and for the hydrogen energy subsystem with slow response speed, the electrochemical reaction model is mainly used to analyze the steady-state and slow dynamic process of temperature and pressure. When overall analysis is needed, joint simulation is usually carried out through simple interface data exchange, and the control strategy is mainly in the form of hierarchical partitioning and independent optimization.

[0004] However, the above analysis method has obvious defects. Due to the huge difference in dynamic response time scale between the electrical subsystem and the hydrogen energy subsystem, the separate modeling and analysis method is difficult to accurately capture the complex, cross-time scale nonlinear coupling effect between the two. In addition, the stability monitoring method based on fixed threshold often has a lag, and usually issues an alarm when the stability has deteriorated significantly. At the same time, the lack of coordinated control strategy may lead to poor control effect or even negative impact when dealing with disturbances. SUMMARY

[0005] To solve the above problems, the application provides a stability analysis and early warning method for a building hydrogen-electricity coupled direct-current microgrid, which can realize the perception and early warning of instability risk and guarantee the stable and reliable operation of the microgrid by fusing the multi-time scale dynamics of the electrical and hydrogen energy subsystems, making forward-looking prediction of the stability evolution trend, and generating a coordinated optimization control strategy.

[0006] The above object can be achieved by the following scheme: A stability analysis and early warning method for a building hydrogen-electricity coupled direct-current microgrid, comprising the following steps: The electromagnetic transient state quantity of the electrical subsystem and the electrochemical reaction state quantity of the hydrogen energy subsystem in the direct current microgrid are acquired, a unified feature matrix of time scale dynamic characteristics is constructed by multi-scale time-frequency decomposition and calculation of cross-domain coupling coefficients; Stability features are extracted from the unified feature matrix in real time, and the stability features are projected to a low-dimensional space through nonlinear mapping to form a representation fingerprint and calculate an instantaneous stability margin under the current state; The representation fingerprint is used for time sequence deduction to calculate the stability evolution trajectory in a future time window and generate a predicted offset vector of the offset direction and amplitude; When the predicted offset vector exceeds the instantaneous stability margin, a collaborative control strategy for coordinating the operating points of the hydrogen energy subsystem and the electrical subsystem is generated with the minimum predicted offset vector as the optimization objective.

[0007] Preferably, the unified feature matrix of time scale dynamic characteristics comprises: The electromagnetic transient state quantity of the electrical subsystem and the electrochemical reaction state quantity of the hydrogen energy subsystem are subjected to multi-scale time-frequency decomposition, respectively, to obtain component features of the electromagnetic transient state quantity and the electrochemical reaction state quantity under different time scales; Based on the intensity and phase relationship of the interaction between the electrical subsystem and the hydrogen energy subsystem in different dynamic processes, cross-domain coupling coefficients between the component features under corresponding time scales are calculated; The component features and the cross-domain coupling coefficients are jointly used as elements to construct the unified feature matrix.

[0008] Preferably, the cross-domain coupling coefficients between the component features under corresponding time scales are calculated by: The asymmetric causal relationship between the component features of the electrical subsystem and the hydrogen energy subsystem under corresponding time scales is identified, and the directional influence intensity is determined; The phase synchronization degree of the component features in the time-frequency domain is calculated to obtain a phase consistency measure of dynamic synergy; The directional influence intensity and the phase consistency measure are fused to generate the cross-domain coupling coefficients representing the action direction, intensity and time sequence synergy relationship.

[0009] Preferably, the representation fingerprint is formed and the instantaneous stability margin under the current state is calculated by: Topological data analysis is performed on the unified feature matrix to construct a geometric manifold in a high-dimensional phase space; Topological invariants of the geometric manifold are extracted, the topological invariants are topological features that exist continuously under different data scales, and the topological invariants are used as elements to form the representation fingerprint; In the low-dimensional space, the shortest distance from the representation fingerprint to the stable boundary of the geometric manifold is calculated and quantified as the instantaneous stability margin.

[0010] Preferably, the geometric manifold includes a stable attractor manifold and an unstable boundary manifold, wherein: The stable attractor manifold is used to characterize the stable operating core domain of the DC microgrid in the high-dimensional phase space, and to depict the inherent recovery capability of returning to the stable equilibrium point under disturbance; The unstable boundary manifold is used to define the critical boundary between the stable attractor manifold and the unstable region, and to provide a criterion for calculating the instantaneous stability margin.

[0011] Preferably, the predicted offset vector generating the offset direction and amplitude includes: Decoupling the characterization fingerprint into electromagnetic dynamic components and electrochemical dynamic components of the electrical subsystem and the hydrogen energy subsystem, respectively; Performing short-time scale deduction on the electromagnetic dynamic components to obtain the electromagnetic transient evolution trend of the fast transient response, and performing long-time scale deduction on the electrochemical dynamic components to obtain the electrochemical steady-state drift trend of the slow variable characteristics; Fusing the electromagnetic transient evolution trend and the electrochemical steady-state drift trend to generate the predicted offset vector.

[0012] Preferably, the cooperative control strategy generating the cooperative regulation of the hydrogen energy subsystem operating point and the electrical subsystem operating point includes: Within the stable attractor manifold, optimization is performed in the opposite direction of the predicted offset vector to determine the target stable point that dissipates the disturbance energy the fastest; Taking the transition cost from the current operating point to the target stable point as the target, an optimization function is constructed, wherein the calculation of the transition cost is weighted in real time by the dynamic coupling weight; By solving the optimization function, the cooperative control strategy of the electrical subsystem and the hydrogen energy subsystem is obtained.

[0013] Preferably, constructing the optimization function includes: Decomposing the transition cost into electrical transition cost and hydrogen energy transition cost required for control effort to migrate from the current operating point to the target stable point; Using the dynamic coupling weight, weight coefficients are assigned to the electrical transition cost and the hydrogen energy transition cost to obtain weighted electrical transition cost and weighted hydrogen energy transition cost; Linearly combining the weighted electrical transition cost and the weighted hydrogen energy transition cost to construct the optimization function.

[0014] Preferably, the method further includes: Based on the cross-domain coupling coefficient and the characterization fingerprint, a joint feature space of the coupling topology is constructed in real time; In the joint feature space, the dynamic coupling weight between the current operating point coupling strength and the topological stability is calculated and extracted.

[0015] The application discloses a stability analysis and early warning system of a building hydrogen-electricity coupling direct-current microgrid, and is used for realizing the method. A unified state construction module is used for acquiring electromagnetic transient state quantities of an electrical subsystem and electrochemical reaction state quantities of a hydrogen energy subsystem in the direct-current microgrid, constructing a unified feature matrix of time scale dynamic characteristics through multi-scale time-frequency decomposition and calculation of cross-domain coupling coefficients, and constructing a unified feature matrix of time scale dynamic characteristics. A stability situation awareness module is used for extracting stability features from the unified feature matrix in real time, projecting the stability features to a low-dimensional space through nonlinear mapping, forming a representation fingerprint, and calculating an instantaneous stability margin under a current state. An evolution trajectory prediction module is used for performing time series deduction by using the representation fingerprint, calculating a stability evolution trajectory in a future time window, and generating a prediction offset vector of an offset direction and amplitude. A forward-looking cooperative control module is used for generating a cooperative control strategy of cooperatively controlling a hydrogen energy subsystem operating point and an electrical subsystem operating point when the prediction offset vector exceeds the instantaneous stability margin, with the optimization objective being to minimize the prediction offset vector.

[0016] Compared with the prior art, the application has the following advantages: The application fuses multi-time scale dynamic information of the electrical and hydrogen energy subsystems, constructs a unified feature matrix capable of comprehensively reflecting inherent coupling characteristics of the system, overcomes the defect of model fragmentation caused by dynamic characteristic differences in traditional analysis, can reveal potential instability modes caused by fast and slow dynamic interaction and difficult to be discovered through independent observation, and thus improves the accuracy and depth of state awareness.

[0017] The application introduces a stability evaluation mechanism based on topological data analysis and a forward-looking evolution trajectory prediction function, abstracts high-dimensional dynamic behavior into a low-dimensional representation fingerprint, quantifies an instantaneous stability margin, realizes robust evaluation of the stability margin, and changes from post-response to pre-warning in combination with a prediction offset vector generated through time series deduction, so that a valuable time window is obtained for taking control measures, and safety and reliability are enhanced.

[0018] The application proposes a cooperative control strategy closely combining stability diagnosis and control decision. The strategy takes a forward-looking prediction result as a trigger condition, takes minimization of an instability trend as an optimization objective, and intelligently allocates control tasks by using dynamic coupling weights, ensures that control resources can be preferentially used to solve a key link most threatening to stability, avoids control conflicts and resource waste, realizes efficient suppression of disturbance and rapid recovery of system stability, and improves autonomous operation and dynamic adjustment capability of the entire microgrid. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1This is a flowchart illustrating the method of the present invention; Figure 2 The evolutionary river map is constructed using a unified feature matrix in Embodiment 1 of this invention; Figure 3 This is a schematic diagram of the state distribution and risk correlation in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the system of the present invention. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0021] Example 1: As Figure 1 As shown, a stability analysis and early warning method for a building hydrogen-electric coupled DC microgrid is proposed. This method employs a fusion characterization of the electrical and hydrogen energy subsystems across multiple time scales, a forward-looking prediction of the stability evolution trend, and the generation of a collaborative optimization control strategy. This enables the perception and early warning of instability risks, ensuring the stable and reliable operation of the microgrid.

[0022] The specific steps of this method include: The electromagnetic transient state quantities of the electrical subsystem and the electrochemical reaction state quantities of the hydrogen energy subsystem in the DC microgrid are obtained. A unified feature matrix of the time-scale dynamic characteristics is constructed by multi-scale time-frequency decomposition and calculation of cross-domain coupling coefficients. Stability features are extracted in real time from a unified feature matrix, and the stability features are projected onto a low-dimensional space through nonlinear mapping to form a characterization fingerprint and calculate the instantaneous stability tolerance under the current state. By using characterization fingerprints for time series extrapolation, the stability evolution trajectory within a future time window is calculated, and a predicted offset vector with offset direction and magnitude is generated. When the predicted offset vector exceeds the instantaneous stability tolerance, a coordinated control strategy is generated to coordinate the operation points of the hydrogen energy subsystem and the electrical subsystem, with the goal of minimizing the predicted offset vector.

[0023] By employing a method of fusion characterization of the electrical and hydrogen energy subsystems across multiple time scales, forward-looking prediction of stability evolution trends, and generation of collaborative optimization control strategies, it is possible to perceive and provide early warning of instability risks, thereby ensuring the stable and reliable operation of the microgrid.

[0024] Constructing a unified feature matrix for time-scale dynamic characteristics includes: Multi-scale time-frequency decomposition was performed on the electromagnetic transient state quantities of the electrical subsystem and the electrochemical reaction state quantities of the hydrogen energy subsystem to obtain the component characteristics of the electromagnetic transient state quantities and the electrochemical reaction state quantities at different time scales. This step aims to decompose the original signals of different physical domains and different variation rates into a unified and comparable framework. The acquisition module obtains the electromagnetic transient state variables of the electrical subsystem and the electrochemical reaction state variables of the hydrogen energy subsystem. For example, the state variable of the electrical subsystem can be the time series data of the DC bus voltage, and the state variable of the hydrogen energy subsystem can be the time series data of the internal temperature or hydrogen pressure of the fuel cell. Subsequently, the data processing module decomposes the two sets of time series data using multi-scale time-frequency analysis tools and continuous wavelet transform. Through this decomposition process, the original one-dimensional time series signal can be mapped to a two-dimensional plane of time and scale, thereby extracting the respective component features at multiple time scales. The number of time scales and the division basis are determined according to the specific hardware composition of the microgrid and the typical disturbance frequency range, with the purpose of ensuring that the key dynamic range from high-frequency switching noise to slow power fluctuations is covered.

[0025] Based on the quantification of the strength and phase relationship of the interaction between the electrical subsystem and the hydrogen energy subsystem in different dynamic processes, the cross-domain coupling coefficient between the component features at each corresponding time scale is calculated. In order to deeply reveal the hidden dynamic correlation between the electrical subsystem and the hydrogen energy subsystem, rather than simply juxtaposing the data, the cross-domain coupling coefficient is introduced. The calculation of this coefficient integrates information from two dimensions: one is the directional influence strength, and the other is the phase consistency measure. The calculation of directional influence strength can utilize the transfer entropy theory in information theory. Transfer entropy can quantify the reduction of uncertainty about the future of one time series by another time series, thereby asymmetrically revealing the influence strength of, for example, the change of electrical component features on the future change of hydrogen energy component features, and vice versa.

[0026] The calculation of phase consistency measure focuses on the degree of coordination of two component features in dynamic changes. This can be achieved by analyzing the synchronization of their instantaneous phases in the time-frequency domain, such as calculating the stability of their phase difference, thereby obtaining a quantitative indicator. Finally, the directional influence strength and the phase consistency measure are integrated, for example, by weighted multiplication, to obtain the cross-domain coupling coefficient that can comprehensively represent the direction, strength, and timing coordination of the relationship.

[0027] The component features and cross-domain coupling coefficients are jointly used as elements to construct a unified feature matrix.

[0028] At each data sampling moment, the electrical component features, hydrogen energy component features and the calculated cross-domain coupling coefficients at all time scales are arranged in order. For example, a row vector can be formed, the structure of which is all electrical component features, all hydrogen energy component features and all cross-domain coupling coefficients in turn. With the passage of time, the continuously collected and calculated row vectors are stacked to form a dynamic evolving time series matrix, that is, the unified feature matrix constructed in this embodiment. Each row of the matrix corresponds to the overall state at a sampling moment, and each column corresponds to a specific feature dimension, providing a highly information-fused data basis for subsequent stability analysis, as shown in Figure 2 Figure 2 In the form of a river map, the dynamic evolution and relative proportion over time of the energy or intensity of key features such as electrical component features, hydrogen energy component features and cross-domain coupling coefficients are shown.

[0029] The calculation of the cross-domain coupling coefficients between the component features at the corresponding time scales includes: Identifying the asymmetric causal relationship between the electrical subsystem and the hydrogen energy subsystem at the corresponding time scales between the component features, and determining the directional influence strength; After obtaining the electrical component features and hydrogen energy component features at each time scale through multi-scale time-frequency decomposition, to identify the asymmetric causal relationship between the two, the transfer entropy theory is used for calculation. The transfer entropy from the electrical component features to the hydrogen energy component features, and the transfer entropy from the hydrogen energy component features to the electrical component features are calculated respectively. The two one-way transfer entropy values quantify the one-way information flow between each other, and the difference or some pre-set combination constitutes the directional influence strength. This strength value not only indicates the strength of the interaction, but more importantly, it can indicate whether the electrical subsystem or the hydrogen energy subsystem is dominant at the current time scale.

[0030] Solving the phase synchronization degree of the component features in the time-frequency domain to obtain a phase consistency measure of dynamic synergy; To solve the synergy of the two component features in dynamic change, the instantaneous phase of the two is extracted. One optional way is to use Hilbert transform, which is a signal processing technique that can convert a real-valued signal into a complex analytic signal, thereby unambiguously obtaining its instantaneous phase information. After obtaining the instantaneous phase of the electrical component features and the hydrogen energy component features respectively, by calculating the difference between the instantaneous phases of the two and analyzing the distribution concentration degree of the phase difference in a short time window, a quantitative phase consistency measure can be obtained. The measure is a normalized value, for example, between 0 and 1. The closer the value is to 1, the higher the synergy of the two component features in dynamic change, that is, the higher the degree of phase synchronization.

[0031] ​The directional influence strength and the phase consistency measure are fused to generate a cross-domain coupling coefficient representing the direction, strength and timing coordination of the effect.

[0032] The directional influence strength and the phase consistency measure are fused to generate a cross-domain coupling coefficient representing the direction, strength and timing coordination of the effect.

[0033] Forming the representation fingerprint and calculating the instantaneous stability margin under the current state include: Topological data analysis is performed on the unified feature matrix to construct a geometric manifold in the high-dimensional phase space. Each row of the matrix can be regarded as a state point in the high-dimensional phase space at a sampling time, and all historical state points collectively form a point cloud data set. In order to reveal the overall dynamic structure hidden behind these discrete data points, topological data analysis techniques are applied to the point cloud data. Topological data analysis is a mathematical method that can extract the intrinsic shape information from data. For example, by constructing a Vietoris-Rips complex, the intrinsic geometric manifold of the dynamic in the phase space can be reconstructed without relying on a specific coordinate system.

[0034] Key topological invariants of the geometric manifold are extracted, and the topological invariants are used as elements to form the representation fingerprint. In order to extract the essential features of the running mode from the complex geometric manifold, a persistent homology algorithm is used to analyze its topological structure. The persistent homology algorithm can identify topological features that persist at different data scales, such as connected components, loops or high-dimensional cavities, etc. These topological features that persist at different scales are considered as key topological invariants. Subsequently, the properties of these topological invariants, such as the birth and death scales of the features, are quantified and combined into a low-dimensional feature vector, which is the representation fingerprint defined in this embodiment. This representation fingerprint describes the overall topological morphology of the current running state in a highly condensed and insensitive manner to data noise.

[0035] In the low-dimensional space, the shortest distance from the representation fingerprint to the stable boundary of the geometric manifold is calculated and quantified as the instantaneous stability margin.

[0036] The stability margin is quantified in a low-dimensional space. The low-dimensional space and the geometric manifold defined therein, which stably bounds the stable region and the unstable region, can be obtained by offline training and learning on a large amount of historical safe operation data and critical instability data. The stable operation region and the unstable region are clearly divided in the geometric manifold. The current time obtained representation fingerprint is mapped to a point in the low-dimensional space, and the shortest distance from the point to the geometric manifold stable boundary is calculated, for example, the shortest Euclidean distance can be calculated. The calculated shortest distance is finally quantified as the instantaneous stability margin, and the numerical value reflects the margin of the current operating state from the instability threshold.

[0037] The geometric manifold includes a stable attractor manifold and an instability boundary manifold, wherein: The stable attractor manifold is used to represent the stable operation core domain of the DC microgrid in the high-dimensional phase space, and to depict the inherent recovery ability under disturbance to return to the stable equilibrium point. This definition is a functional division for constructing the geometric manifold. The division process is based on a large-scale data set composed of historical state points and simulation state points in a unified feature matrix. To make the division, each state point in the data set needs to be labeled first. The basis for labeling is the dynamic evolution behavior after the state point: if it can recover to equilibrium after being disturbed slightly, the point is labeled as stable; if it evolves to voltage collapse or sustained oscillation, the point is labeled as unstable. Based on this labeling, all the state points labeled as stable are collected together, and the intrinsic geometric structure of them in the high-dimensional phase space is reconstructed using manifold learning algorithm. The structure is the stable attractor manifold. This manifold exhibits as a collection area of all normal operating states in shape, and the most stable equilibrium point corresponding to the geometric center or the highest density area of the manifold, and the volume and smoothness of the manifold depict the inherent resilience against disturbance and self-recovery.

[0038] The instability boundary manifold is used to define the critical boundary between the stable attractor manifold and the unstable region, and to provide a criterion for calculating the instantaneous stability margin.

[0039] The construction of the instability boundary manifold uses all the labeled data points. By training a classification model, such as a support vector machine, a dividing surface that can best separate the stable point set and the unstable point set in the high-dimensional phase space is found. This dividing surface itself is a high-dimensional surface, which is the instability boundary manifold. This manifold defines the edge of the stable region, and any state trajectory that crosses this boundary will be judged to enter the instability process. Therefore, the instability boundary manifold provides a clear geometric criterion for the calculation of the instantaneous stability margin, that is, the distance of the point corresponding to the current state representation fingerprint to the boundary.

[0040] The predicted offset vector including generating an offset direction and amplitude includes: decoupling the characterization fingerprint into electromagnetic dynamic component and electrochemical dynamic component respectively; The characterization fingerprint obtained in real time is taken as input. The vector structure of the characterization fingerprint retains the correspondence between its elements and the original physical subsystems when it is constructed. Therefore, the characterization fingerprint, which is a low-dimensional vector, is divided into two parts, thereby decoupling to obtain the electromagnetic dynamic component and the electrochemical dynamic component reflecting the dynamic characteristics of the two subsystems respectively.

[0041] The electromagnetic dynamic component is deduced in short time scale to obtain the electromagnetic transient evolution trend of fast transient response, and the electrochemical dynamic component is deduced in long time scale to obtain the electrochemical steady-state drift trend of slow varying characteristics; This step adopts a double time scale prediction strategy. For the electromagnetic dynamic component, a short-term prediction model is used, such as a long short-term memory network, which is specially trained to learn the fast dynamic behavior in milliseconds to seconds. The sequence of electromagnetic dynamic component in the past period of time is taken as input, and the network can output its prediction value in the future short time window, and the difference between the prediction value and the current value constitutes the electromagnetic transient evolution trend reflecting the fast transient response. Similarly, for the electrochemical dynamic component, a long-term prediction model is used, which is specially trained to learn the slow varying characteristics by sampling and training in minutes or longer time scale. The sequence of electrochemical dynamic component in the past longer time is taken as input, and the model can output its prediction value in the future long time window, and the difference between the prediction value and the current value constitutes the electrochemical steady-state drift trend reflecting the slow varying characteristics.

[0042] The electromagnetic transient evolution trend and the electrochemical steady-state drift trend are fused to generate a prediction offset vector.

[0043] To obtain the comprehensive evolution direction and amplitude in the future time window, the two trend vectors obtained in the previous step need to be fused. One optional way is to perform weighted fusion. The setting of the weighting coefficient is based on the different influence degrees of the electrical subsystem and the hydrogen energy subsystem in the overall stability, and its specific value can be determined by offline simulation sensitivity analysis or optimization algorithm to adjust the relative importance of fast and slow dynamics in the final prediction. The result of fusion is the final prediction offset vector, which indicates the most likely evolution direction and amplitude of the state in the future time window in the low-dimensional feature space.

[0044] Generating a collaborative control strategy for coordinating the operation points of the hydrogen energy subsystem and the electrical subsystem includes: Within the stable attractor manifold, optimization is performed in the opposite direction of the prediction offset vector to determine the target stable point that can dissipate the disturbance energy fastest; The optimization process is conducted in the low-dimensional representation space of the geometric manifold. The process starts from the current operating point corresponding to the representative fingerprint position, and is strictly constrained within the safe region of the stable attractor manifold. The optimization direction is searched in the opposite direction of the predicted offset vector. The selection criteria for this point not only deviates from the current unstable trend, but also meets the characteristics of the fastest energy dissipation of the disturbance, for example, it can correspond to the fastest gradient descent direction in the stable attractor manifold. The result obtained by this way of optimization is the target stable point.

[0045] The optimization function is constructed to minimize the transition cost from the current operating point to the target stable point, and the calculation of the transition cost is weighted in real time by the dynamic coupling weight; This step constructs an optimization function to minimize the transition cost from the current operating point to the target stable point. The transition cost here is not simply a geometric distance, but quantifies the control cost required to drive the state from the current point to the target point. In calculating this transition cost, it will be weighted in real time by the dynamic coupling weight. This means that if the coupling effect on a certain time scale is identified as the main threat to the current stability, the cost of the control action required to suppress the coupling effect will be given higher priority or more important consideration in the optimization function.

[0046] By solving the optimization function, the coordinated control strategy of the electrical subsystem and the hydrogen energy subsystem is obtained.

[0047] By solving the above constructed weighted optimization function, a set of optimal control variable adjustment amounts can be obtained. These adjustment amounts are then mapped back to the specific controllable devices of the electrical subsystem and the hydrogen energy subsystem, such as adjusting the power reference value of the converter in the DC microgrid, or adjusting the operating current of the electrolyzer in the hydrogen energy subsystem. This set of specific, executable device setting value adjustment instructions constitutes the final coordinated control strategy, which proactively avoids the impending instability risk, such as Figure 3 shown, Figure 3 The negative correlation between the cross-domain coupling strength and the instantaneous stability margin is shown in the form of scatter plot and edge histogram, and the distribution area of high-risk state points is revealed.

[0048] The construction of the optimization function includes: The transition cost is decomposed into the electrical transition cost and the hydrogen energy transition cost required for the control effort to migrate from the current operating point to the target stable point; After the target steady state is determined, the control adjustment needed to transfer from the current operating point to the target steady state is quantified. The overall transition cost is decomposed into two independent parts. The first part is the electrical transition cost, which quantifies the control effort or energy consumption needed to adjust the electrical subsystem control variables to reach the target state, such as adjusting the photovoltaic inverter output power or the energy storage charge / discharge rate. The second part is the hydrogen transition cost, which quantifies the corresponding cost needed to adjust the hydrogen subsystem control variables, such as adjusting the electrolyzer input current or the fuel cell hydrogen production rate. Both cost terms are functions of the respective control variable adjustment amounts.

[0049] The dynamic coupling weights are utilized to assign weight coefficients to the electrical transition cost and the hydrogen transition cost, resulting in a weighted electrical transition cost and a weighted hydrogen transition cost; The dynamic coupling weights calculated in real time are utilized to dynamically assign weight coefficients to the aforementioned two cost terms. The dynamic coupling weights related to the fast dynamics of the electrical subsystem are aggregated to form the electrical weight coefficient; the dynamic coupling weights related to the slow dynamics of the hydrogen subsystem are aggregated to form the hydrogen weight coefficient. Subsequently, these coefficients are multiplied by the corresponding cost terms to obtain the weighted electrical transition cost and the weighted hydrogen transition cost.

[0050] The weighted electrical transition cost and the weighted hydrogen transition cost are linearly combined to construct an optimization function.

[0051] The two weighted cost terms are linearly combined to construct the final optimization function that needs to be solved, which is expressed as: ; wherein, is the overall optimization function that needs to be minimized at time t, and the solution of which will guide the generation of the control strategy; and are the electrical weight coefficient and the hydrogen weight coefficient at time t, respectively, which are derived from the aggregation of the calculated dynamic coupling weights and are real-time values that dynamically change over time, reflecting the different influence degrees of the electrical and hydrogen subsystems on the overall stability at the current time; is the electrical transition cost, which is derived from the electrical control adjustment amount , which is used to represent the control cost; is the hydrogen transition cost, which is derived from the hydrogen control adjustment amount , and represent the control variable adjustment amounts needed for the electrical subsystem and the hydrogen subsystem, respectively, to transfer from the current operating point to the target steady state.

[0052] Embodiment 2: Based on Embodiment 1, the method further comprises: Based on the cross-domain coupling coefficients and the characterization fingerprints, a joint feature space of the coupling topology is constructed in real time; This step starts with real-time acquisition of the generated cross-domain coupling coefficients and the characterization fingerprints. The cross-domain coupling coefficients are a feature vector, whose elements characterize the interaction strength and direction between the electrical subsystem and the hydrogen energy subsystem at different time scales; while the characterization fingerprints are another feature vector, which summarizes the overall stability of the current operating state from the topology perspective. This method first combines these two feature vectors of different origins but synchronized in time, for example, by vector splicing, to form a higher-dimensional joint feature vector. The space formed by all these joint feature vectors is the joint feature space of the coupling topology, which contains comprehensive information about the internal mechanism and external macroscopic performance at each moment.

[0053] In the joint feature space, the dynamic coupling weights between the coupling strength and the topology stability of the current operating point are calculated and extracted.

[0054] To extract the dynamic coupling weights, a mapping model from the joint feature space to the stability scalar is needed to be established first. The input of this model is the joint feature vector constructed in the previous step, and the output is a scalar that can directly measure the stability degree. This stability scalar can be the instantaneous stability margin calculated in the previous step. This mapping model can be established by learning methods such as support vector regression machine, using historical data for offline training. In real-time operation, the dynamic coupling weights are obtained by calculating the partial derivative of the stability scalar with respect to each cross-domain coupling coefficient component in the joint feature vector. The calculation process can be represented as: ; where, is the dynamic coupling weight corresponding to time scale j at time t, which is a set of weight values for the subsequent control strategy; is the scalar representing stability, which is derived from the instantaneous stability margin calculated in the previous step, and is a known real-time value; is the cross-domain coupling coefficient corresponding to time scale j at time t, which is derived from the real-time feature value calculated by the method. The absolute value of this partial derivative quantifies the sensitivity of the overall stability to the small change in the cross-domain coupling strength at a specific time scale under the current state, which together constitute the dynamic coupling weights of the current operating point.

[0055] To verify the feasibility and effectiveness of the method in practical application, the method is applied to a digital twin simulation platform of a building hydrogen-electricity coupled direct current microgrid. The microgrid system is mainly composed of photovoltaic array, lithium battery energy storage, PEM proton exchange membrane electrolyzer and PEM fuel cell, aiming to provide high reliability direct current power supply for an office building and internal data center. The difficulty of this scenario is that the response of the electrical subsystem is in milliseconds, while the electrochemical reaction dynamics of the hydrogen energy subsystem is in minutes, and the complex coupling between the two brings challenges to the stability that traditional methods are difficult to cope with.

[0056] In the verification, a typical running scenario is simulated: during the stable operation of the microgrid, the performance of a key component in the hydrogen energy subsystem slowly deteriorates, and then the electrical subsystem encounters a rapid transient impact.

[0057] In the initial stage of verification, the microgrid is stably running. The method obtains the time series data of representative electrical state quantities and electrochemical state quantities in real time. Through continuous wavelet transform, the signal is decomposed into multiple time scales to obtain electrical component features and hydrogen energy component features at each scale. Then, by calculating the transfer entropy and instantaneous phase synchronization between the two, the cross-domain coupling coefficients are generated, and the unified feature matrix is constructed. In this stable stage, the cross-domain coupling coefficients at each time scale are at a low level, indicating weak coupling between the subsystems.

[0058] After several hours of simulation, the slow decline in heat dissipation efficiency of the fuel cell due to aging is simulated, which is manifested as the gradual drift of its key temperature to the edge of the normal working interval within several hours. Based on the real-time updated unified feature matrix, the method reconstructs the geometric manifold in high-dimensional phase space through topological data analysis technology. The topological invariants of the manifold are extracted through persistent homology algorithm to generate the representation fingerprint.

[0059] According to this, the shortest distance from the current representation fingerprint to the pre-learned unstable boundary manifold, i.e. the instantaneous stability margin, is calculated. The data shows that within the hours of temperature drift, the instantaneous stability margin presents a continuous and significant decline from a high safety value, indicating that the stability margin has been greatly reduced. It is worth noting that during the entire slow degradation process, key electrical physical quantities such as the direct current bus voltage have not shown obvious abnormalities.

[0060] At the same time, by constructing the joint feature space of coupled topology and calculating the dynamic coupling weight in real time, it is found that the dynamic coupling weight corresponding to the minute-level slow dynamic increases significantly at this stage. This successfully attributes the decline in stability margin to the slow parameter drift of the hydrogen energy subsystem, locating the main threat to stability.

[0061] At this moment, the simulation system introduces a fast transient event: the photovoltaic output drops sharply within seconds due to cloud cover. The evolution trajectory prediction of the method of the embodiment is started immediately, decouples the current representation fingerprint into electromagnetic dynamic components and electrochemical dynamic components, and inputs them into short-time and long-time prediction models respectively. The fused result generates a prediction offset vector, which not only contains the fast evolution trend triggered by the photovoltaic sharp drop, but also superimposes the long-term influence of the slow drift of the fuel cell temperature.

[0062] It is calculated that the amplitude of the prediction offset vector has exceeded the current instantaneous stability margin that has been reduced, and the highest level of warning is triggered immediately.

[0063] After the warning is triggered, the generation process of the cooperative control strategy is activated. First, an optimal target stable point is searched in the opposite direction of the prediction offset vector within the stable attractor manifold. Then, an optimization function with the transfer cost as the target is constructed. Since the dynamic coupling weight has identified that the hydrogen energy subsystem is the main contradiction, the weight coefficient related to the hydrogen energy transfer cost in the optimization function is dynamically adjusted to be higher.

[0064] After solving the optimization function, the generated cooperative control strategy is not a single high-power compensation command for the energy storage system, but a combined strategy: first, the energy storage system outputs medium power to quickly support the bus voltage; second, the output power instruction of the fuel cell is appropriately reduced to reduce its thermal load; third, the working state of the electrolyzer is slightly adjusted to assist in balancing.

[0065] After the strategy is executed, the state is successfully guided back to the central region of the stable attractor manifold, and the DC bus voltage fluctuation is effectively suppressed, avoiding a voltage collapse event that may be caused by the superposition of fast and slow dynamics. As a comparison, under the traditional control strategy without the embodiment, the same fault scenario leads to instability due to chain reaction.

[0066] Table 1 State evolution and stability evaluation Table 2 Control strategy and effect comparison From the above verification process and the data in Table 1 and Table 2, it can be seen that the method of the embodiment can effectively monitor the deep stability of the system. Table 1 shows that, in the case where there is no obvious abnormality in the physical quantity, the method of the embodiment successfully quantifies the continuous decrease of the stability margin caused by the slow parameter drift through the instantaneous stability margin index, and successfully locates the risk source. The comparison results in Table 2 reflect the effect of the cooperative control strategy of the method of the embodiment, which resolves the complex instability risk caused by the coupling of multi-time scale dynamics through dynamic weighting and cooperation, and its effect is better than that of the traditional independent control method.

[0067] Embodiment 3: as shown in Figure 4 A stability analysis and early warning system for a building hydrogen-electricity coupled direct current microgrid, comprising: A unified state construction module for obtaining electromagnetic transient state quantities of electrical subsystems and electrochemical reaction state quantities of hydrogen energy subsystems in the direct current microgrid, constructing a unified feature matrix of time scale dynamic characteristics through multi-scale time-frequency decomposition and calculation of cross-domain coupling coefficients; A stability situation awareness module for real-time extraction of stability features from the unified feature matrix, projection of the stability features to a low-dimensional space through nonlinear mapping, formation of a characterization fingerprint, and calculation of an instantaneous stability margin under the current state; An evolutionary trajectory prediction module for time series deduction using the characterization fingerprint, calculation of a stability evolution trajectory within a future time window, and generation of a prediction offset vector of the offset direction and amplitude; A forward-looking collaborative control module for generating a collaborative control strategy for collaborative regulation of the hydrogen energy subsystem operating point and the electrical subsystem operating point when the prediction offset vector exceeds the instantaneous stability margin, with the optimization objective being minimization of the prediction offset vector.

Claims

1. A stability analysis and early warning method for a building hydrogen-electricity coupled direct current microgrid, characterized in that, The method comprises the following steps: acquiring electromagnetic transient state quantities of an electrical subsystem and electrochemical reaction state quantities of a hydrogen energy subsystem in a direct current microgrid, constructing a unified feature matrix of time scale dynamic characteristics through multi-scale time-frequency decomposition and calculation of cross-domain coupling coefficients; real-time extraction of stability features from the unified feature matrix, projection of the stability features to a low-dimensional space through nonlinear mapping to form a representation fingerprint and calculate an instantaneous stability margin under the current state; time sequence deduction using the representation fingerprint to calculate a stability evolution trajectory within a future time window and generate a prediction offset vector of the offset direction and amplitude; when the prediction offset vector exceeds the instantaneous stability margin, generating a collaborative control strategy for collaborative regulation of the hydrogen energy subsystem operating point and the electrical subsystem operating point as an optimization objective of minimizing the prediction offset vector.

2. The stability analysis and early warning method of a building hydrogen-electricity coupled direct current microgrid according to claim 1, characterized in that, The construction of the unified feature matrix of time scale dynamic characteristics comprises: multi-scale time-frequency decomposition of the electromagnetic transient state quantities of the electrical subsystem and the electrochemical reaction state quantities of the hydrogen energy subsystem to obtain component features of the electromagnetic transient state quantities and the electrochemical reaction state quantities under different time scales; calculation of cross-domain coupling coefficients between the component features under each corresponding time scale based on the strength and phase relationship of the interaction between the electrical subsystem and the hydrogen energy subsystem in different dynamic processes; the component features and the cross-domain coupling coefficients are jointly used as elements to construct the unified feature matrix.

3. The stability analysis and early warning method of a building hydrogen-electricity coupled direct current microgrid according to claim 2, characterized in that, The calculation of the cross-domain coupling coefficients between the component features under each corresponding time scale comprises: identification of the asymmetric causal relationship between the component features of the electrical subsystem and the hydrogen energy subsystem under each corresponding time scale to determine the directional influence strength; calculation of the phase synchronization degree of the component features in the time-frequency domain to obtain a phase consistency measure of dynamic collaboration; fusion of the directional influence strength and the phase consistency measure to generate a cross-domain coupling coefficient representing the action direction, strength and time sequence collaboration relationship.

4. The stability analysis and early warning method of a building hydrogen-electricity coupled direct current microgrid according to claim 1, characterized in that, The formation of the representation fingerprint and the calculation of the instantaneous stability margin under the current state comprises: topological data analysis of the unified feature matrix to construct a geometric manifold in a high-dimensional phase space; extraction of topological invariants of the geometric manifold, the topological invariants being topological features that exist continuously under different data scales, and the topological invariants being used as elements to form the representation fingerprint; calculation of the shortest distance from the representation fingerprint to the stable boundary of the geometric manifold in the low-dimensional space and quantification as the instantaneous stability margin.

5. The stability analysis and early warning method of a building hydrogen-electricity coupled direct current microgrid according to claim 4, characterized in that, The geometric manifold comprises a stable attractor manifold and an unstable boundary manifold, wherein: the stable attractor manifold is used to represent the stable operation core domain of the direct current microgrid in the high-dimensional phase space and to depict the inherent recovery ability of returning to the stable equilibrium point under disturbance; the unstable boundary manifold is used to define the critical boundary between the stable attractor manifold and the unstable region and to provide a criterion for calculating the instantaneous stability margin.

6. The stability analysis and early warning method of a building hydrogen-electricity coupled direct current microgrid according to claim 1, characterized in that, The generation of the prediction offset vector of the offset direction and amplitude comprises: decoupling of the representation fingerprint into electromagnetic dynamic components and electrochemical dynamic components of the electrical subsystem and the hydrogen energy subsystem, respectively; short time scale deduction of the electromagnetic dynamic components to obtain the electromagnetic transient evolution trend of the fast transient response and long time scale deduction of the electrochemical dynamic components to obtain the electrochemical steady state drift trend of the slow varying characteristics; The prediction offset vector is generated by fusing the electromagnetic transient evolution trend and the electrochemical steady-state drift trend.

7. The stability analysis and early warning method of a building hydrogen-electricity coupled direct current microgrid according to claim 5, characterized in that, The cooperative control strategy for jointly regulating the operation point of the hydrogen energy subsystem and the operation point of the electrical subsystem includes: Within the stable attractor manifold, optimization is performed in the opposite direction of the prediction offset vector to determine the target stable point that dissipates the disturbance energy fastest; An optimization function is constructed with the transition cost from the current operation point to the target stable point as the target, and the calculation of the transition cost is weighted in real time by the dynamic coupling weight; The cooperative control strategy for the electrical subsystem and the hydrogen energy subsystem is obtained by solving the optimization function.

8. The stability analysis and early warning method of a building hydrogen-electricity coupled direct current microgrid according to claim 7, characterized in that, The optimization function includes: The transition cost is decomposed into an electrical transition cost and a hydrogen energy transition cost required for the control effort to migrate from the current operation point to the target stable point; The dynamic coupling weight is used to assign weight coefficients to the electrical transition cost and the hydrogen energy transition cost to obtain a weighted electrical transition cost and a weighted hydrogen energy transition cost; The weighted electrical transition cost and the weighted hydrogen energy transition cost are linearly combined to construct the optimization function.

9. The stability analysis and early warning method of a building hydrogen-electricity coupled direct current microgrid according to claim 1, characterized in that, The method further includes: Based on the cross-domain coupling coefficient and the fingerprint, a joint feature space of the coupling topology is constructed in real time; In the joint feature space, the dynamic coupling weight between the current operation point coupling strength and the topology stability is calculated and extracted.

10. A stability analysis and early warning system for a building hydrogen-electricity coupled direct current microgrid, configured to implement the method of any one of claims 1-9. It includes: A unified state construction module is used to obtain the electromagnetic transient state quantity of the electrical subsystem and the electrochemical reaction state quantity of the hydrogen energy subsystem in the DC microgrid, construct a unified feature matrix of time scale dynamic characteristics through multi-scale time-frequency decomposition and calculation of cross-domain coupling coefficients, and obtain the instantaneous stability margin under the current state by nonlinear mapping to form a fingerprint and calculate the stability feature. A stability situation awareness module is used to extract stability features in real time from the unified feature matrix, project the stability features to a low-dimensional space through nonlinear mapping, form a fingerprint, and calculate the instantaneous stability margin under the current state; An evolution trajectory prediction module is used to perform time series inference using the fingerprint to calculate the stability evolution trajectory in the future time window and generate a prediction offset vector of the offset direction and amplitude; A forward-looking cooperative control module is used to generate a cooperative control strategy for jointly regulating the operation point of the hydrogen energy subsystem and the operation point of the electrical subsystem when the prediction offset vector exceeds the instantaneous stability margin, with the optimization objective being to minimize the prediction offset vector.

Images