Optimization method for multi-source energy storage operation performance and fault state characterization parameters

Abstract

The invention discloses an optimization method for multi-source energy storage operation performance and fault state characterization parameters, and the method comprises the steps: S1, carrying out the synchronous collection and coupling mapping of multi-domain signals, constructing a coupling function family based on an energy conservation law, mapping heterogeneous signals, such as voltage, rotating speed, temperature difference and the like, to a unified high-dimensional space, and outputting inter-domain coupling gain factors; s2, constructing an asymmetric disturbance response spectrum, and extracting parameters such as response asymmetry to form a fault sensitive fingerprint; s3, taking the fingerprint as input, driving a sensing topology ad hoc network through an evolutionary game, and dynamically generating an elastic topology; s4, game equilibrium points are extracted, an equilibrium-robust response mapping model is established, and a robust threshold value is output; s5, parameters are dynamically adjusted based on game residual errors, and the topology is corrected in a zero-delay mode; and S6, calculating a game convergence entropy, and forming a health early warning closed loop. According to the method, full-life-cycle health management of the multi-source energy storage system is realized, and the fault sensing sensitivity and the system robustness are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of multi-source energy storage system state monitoring and safety protection, and particularly relates to an optimization method for multi-source energy storage operation performance and fault state characterization parameters. BACKGROUND

[0002] Current situation of prior art:

[0003] In the current multi-source energy storage system, devices such as electrochemical energy storage, inertia energy storage and thermal energy storage usually adopt independent monitoring methods, that is, single type sensors are deployed to collect heterogeneous signals such as voltage, speed and temperature difference, and then traditional centralized or distributed networks are used for data transmission and analysis. The prior art relies on single domain signal features for fault judgment, such as judging the electrochemical energy storage state by battery voltage fluctuation or evaluating the inertia energy storage performance by flywheel speed change. The network topology is mostly fixed structure, which is difficult to adapt to the dynamic changes of the system.

[0004] Defects of prior art:

[0005] The prior art has three significant defects: first, the heterogeneous signals have poor coupling, the cross-domain signals such as voltage, speed and temperature difference lack unified characterization dimension, and it is difficult to reflect the cooperative fault characteristics of the multi-source energy storage system; second, the network topology has poor adaptability, the fixed topology cannot be dynamically adjusted in the early stage of information loss or fault, resulting in low fault sensing sensitivity and waste of communication resources; third, the state evaluation and early warning closed loop is missing, the existing technology mostly stays in single fault parameter monitoring, lacks the whole process closed loop management from "sensing-evaluation-early warning-optimization", and is difficult to cope with non-stationary conditions such as system aging and load mutation (the above defects are directly solved by the technical scheme of the present application).

[0006] Therefore, we propose an optimization method for multi-source energy storage operation performance and fault state characterization parameters. SUMMARY

[0007] In view of the above defects in the prior art, the purpose of the present application is to solve the problems of poor cooperation of multi-source energy storage heterogeneous signals, poor topology adaptability and lack of closed loop management, and to realize the whole life cycle health management and robust optimization.

[0008] In order to achieve the above purpose, the technical scheme of the present application is as follows:

[0009] The optimization method for multi-source energy storage operation performance and fault state characterization parameters comprises the following steps:

[0010] 1. The optimization method for multi-source energy storage operation performance and fault state characterization parameters, characterized in that it comprises the following steps:

[0011] S1. Multi-domain signal synchronous acquisition and coupling mapping, acquire voltage signal, speed signal, temperature difference signal, construct coupling function family based on energy conservation law, map heterogeneous signals to unified high-dimensional state space, output inter-domain coupling gain factor;

[0012] S2. Asymmetric disturbance response spectrum construction, inject controllable asymmetric perturbation signals in the unified high-dimensional state space, construct a closed-loop disturbance-response experiment framework, extract response asymmetry, delay coupling coefficient and dissipation offset, and form a unique fault sensitive fingerprint after normalization processing;

[0013] S3. Evolutionary game driven perception topology self-organization, take the fault sensitive fingerprint as input, deploy distributed perception agents at key nodes of the energy storage system; introduce strategy drift rate and information compensation factor as game variables, evolve connection strategy under information missing condition, and dynamically generate elastic topology;

[0014] S4. Game equilibrium-robust response mapping optimization, extract the converged equilibrium point from the evolutionary game process of the elastic topology, establish an equilibrium-robust response mapping model, and inversely solve the optimal network configuration through multi-perturbation simulation to output reusable robust threshold;

[0015] S5. Residual driven online updating mechanism, real-time calculate the deviation between the current connection strategy of the distributed perception agent and the game equilibrium point, dynamically adjust the weights of the strategy drift rate and the information compensation factor using the residual-delay mapping function, and realize zero-delay correction of the perception topology and operating parameters;

[0016] S6. Game convergence entropy health warning closed loop, calculate the game convergence entropy according to the convergence speed and stability of the distributed perception agent, compare the game convergence entropy with the reusable robust threshold in real time, and trigger different levels of warning according to the degree of exceeding the threshold.

[0017] Further, the voltage signal is collected through an electrochemical energy storage module, the speed signal is collected through an inertial energy storage unit, and the temperature difference signal is collected through a thermal energy storage device; The acquisition process adopts Beidou time service to ensure time synchronization, with a synchronization error controlled within 10μs, ensuring the consistency of different domain signals in time dimension.

[0018] Further, the step of constructing a coupling function family based on the law of conservation of energy, mapping heterogeneous signals to a unified high-dimensional state space, and outputting inter-domain coupling gain factor comprises:

[0019] S111. Based on the law of conservation of energy, analyze the conversion relationship among the electrical energy of electrochemical energy storage, the kinetic energy of inertial energy storage, and the thermal energy of thermal energy storage, construct a coupling function family containing multiple mapping relationships such as voltage-speed, speed-temperature difference, and voltage-temperature difference, to realize the correlation and quantification of different energy signals;

[0020] The core expression of the coupling function family is the total energy coupling function, which has the following form:

[0021] ;

[0022] is the value of the total energy coupling function; electrical energy of electrochemical energy storage; kinetic energy of inertia energy storage; thermal energy of thermal energy storage; is the weight coefficient in the domain, satisfying ; is the inter-domain coupling coefficient, which quantifies the interaction strength between different domain energies;

[0023] S112. Heterogeneous signal mapping, the collected heterogeneous signals are respectively converted into energy parameters of the corresponding domain, and then are mapped to a unified high-dimensional state space with a dimension of 10-20 dimensions through coordinate transformation by the coupling function family, to realize standardized expression of the heterogeneous signals;

[0024] S113. Output of inter-domain coupling gain factor, in the process of mapping the heterogeneous signals to the high-dimensional state space, the inter-domain coupling gain factor is calculated and output in real time, which quantifies the mutual influence strength between different domain signals, and provides a consistent coordinate system and cross-domain correlation benchmark for subsequent fault sensitive analysis; the calculation formula of the inter-domain coupling gain factor is:

[0025] ;

[0026] is the inter-domain coupling gain factor; respectively represent any two domains in electrochemical, inertia, and thermal energy storage; The larger the value of , the stronger the coupling effect between the domain signal.

[0027] Further, the asymmetric perturbation is defined as:

[0028] ;

[0029] is the perturbation signal in the th dimension of the high-dimensional state space; , are the amplitudes of the forward and reverse perturbations, respectively, ; , are the action times of the forward and reverse perturbations, respectively, ; is the perturbation angular frequency, which maintains a certain proportion with the inherent frequency of the system to excite significant response;

[0030] And the controllable asymmetric perturbation amplitude is 0.1%-5% of the normal signal amplitude, and has asymmetry in the positive and negative directions and the action time.

[0031] Further, the experimental framework comprises a perturbation generator, a high-dimensional state monitoring module, a response signal acquisition module, and a feedback control unit. The experimental framework ensures that the perturbation is injected according to the preset strategy, and the response signal of the system under the action of the perturbation is acquired in real time and synchronously:

[0032] The response asymmetry quantification system measures the difference in response to forward and reverse asymmetric perturbations. The greater the difference, the more significant the fault precursor. The calculation formula is:

[0033] ;

[0034] The response asymmetry is represented by the value, and the greater the value, the more significant the difference in response to asymmetric perturbations by the system;

[0035] is the high-dimensional space dimension; is the forward perturbation response peak value; is the reverse perturbation response peak value; is the reverse perturbation response peak value is the corresponding response time; is the forward perturbation response peak value is the corresponding response time;

[0036] The delay coupling coefficient represents the time delay and correlation strength between the perturbation signal and the response signal, and reflects the abnormality of the dynamic characteristics of the system. The calculation formula is:

[0037] ;

[0038] is the delay coupling coefficient; is the perturbation signal vector, which includes the perturbation values of each dimension in the high-dimensional space;

[0039] is the response signal vector, which includes the response values of each dimension in the high-dimensional space; is the delay time; 、 are the mean vectors of the perturbation and the response, respectively, used to eliminate the influence of the direct current component;

[0040] The dissipation offset measures the deviation of the energy dissipation of the system under the action of the perturbation from the normal state, and reflects the abnormality of the energy conversion efficiency. The calculation formula is:

[0041] ;

[0042] is the dissipation offset, The larger the value, the more obvious the abnormal energy dissipation of the system (such as the failure of the heat storage insulation layer, Will significantly increase); is the total perturbation time; is the L2 norm, which represents the energy size of the perturbation input; The L2 norm of the response signal represents the energy size of the system response.

[0043] Further, the fault fingerprint is used as the core feature to distinguish different fault types, and is used to quantify the sensitive differences of different fault precursors. The fault sensitive fingerprint is represented as:

[0044] ;

[0045] ; ; ;

[0046] wherein, is the normalized response asymmetry; is the normalized delay coupling coefficient; is the normalized dissipation offset.

[0047] Further, the distributed sensing agent has the ability to collect local state signals of the node in real time, process local data based on built-in algorithms, and interact with adjacent agents through wireless or wired methods. At the same time, the initial computing resources, communication bandwidth and storage capacity of the distributed sensing agent are allocated to ensure that it can carry out game calculation and information interaction tasks;

[0048] The strategy drift rate represents the probability of the distributed sensing agent changing the current connection strategy, reflecting the flexibility of topology adjustment. Its value is related to the fault sensitive fingerprint. When the fault sensitive fingerprint fluctuates greatly, the drift rate is in the range of 0.3-0.5, which promotes the distributed sensing agent to actively adjust the connection. When the fault sensitive fingerprint is stable, the drift rate decreases to 0.1-0.2, maintaining the stability of the topology.

[0049] The information compensation factor is used to make up for the lack of information caused by communication delay or signal loss. The value range is 0.5-1.0. When the information loss degree is > 30%, the factor value is increased to 0.8-1.0, and the missing data is compensated by information sharing of adjacent distributed sensing agents. When the information integrity is ≥ 90%, the factor decreases to 0.5-0.7, reducing redundant calculation.

[0050] Further, the evolution of the connection strategy under the condition of information loss and the dynamic generation of the elastic topology include the following steps:

[0051] S311. Game rule initialization and benefit function construction:

[0052] Rule setting: the core game rule of the distributed sensing agent is to maximize information benefit and minimize communication cost. Strategy evaluation is performed every 50-100 ms; the evaluation period is dynamically adjusted according to the fluctuation degree of the fault sensitive fingerprint: when > 0.5, it is shortened to 50 ms to quickly respond to faults; when < 0.2, it is extended to 100 ms to reduce resource consumption.

[0053] Benefit function construction: based on the functional characteristics of the distributed sensing agent, the benefit function is constructed as the basis for strategy adjustment: ; wherein, is the weight coefficient, increases with the increase of , and ; the information integrity is the proportion of the effective fault sensitive fingerprint parameters obtained by the distributed sensing agent; the communication cost is positively correlated with the number of connected nodes; is the information compensation factor, and the missing information estimate value is derived from the shared data of adjacent agents;

[0054] S312. Connection strategy evolution:

[0055] Strategy adjustment trigger: according to the current strategy drift rate and the information compensation factor , the agent decides whether to adjust the connection strategy:

[0056] When the current strategy drift rate is 0.3-0.5, there is a 30%-50% probability of triggering strategy adjustment every round of evaluation; when the current strategy drift rate is 0.1-0.2, the adjustment probability is reduced to 10%-20%, reducing unnecessary topology changes.

[0057] Information missing processing: if the information missing degree is > 30%, the distributed sensing agent preferentially sends information request to adjacent nodes, and the calculation formula of the information sharing weight is: ;

[0058] The higher the value of the information sharing weight of a node, the higher the priority of receiving information sharing request, ensuring that missing data is efficiently compensated;

[0059] S313. Elastic topology expansion: when the benefit function , topology expansion is triggered:

[0060] Expansion logic: the distributed sensing agent increases the connection with high information value nodes,

[0061] Expansion quantity calculation formula: ;

[0062] Wherein, is the expansion coefficient, is the rounding function; is the current strategy drift rate;

[0063] Expansion target, preferentially connecting fault-sensitive fingerprint complementary nodes, expanding the topology diameter from the initial 3-4 hops to 5-6 hops, covering a wider fault perception range;

[0064] S314. Elastic topology contraction, when the benefit function U<0.3, trigger topology contraction:

[0065] Distributed perception agent disconnects with low information value nodes, and the number of connections

[0066] Calculation formula: ;

[0067] Wherein, is the contraction coefficient;

[0068] Contraction target, preferentially disconnecting connections with information repetition rate >80%, shrinking the topology diameter to the initial level, and reducing communication energy consumption;

[0069] S315. Topology stability verification and iteration:

[0070] Stability verification, after each expansion or contraction, calculate the connection stability index :

[0071] ;

[0072] When ≥0.8, it is determined that the topology enters a stable state; if <0.8, repeat steps S312-S314 until the topology converges;

[0073] Iterative update, as the system state changes, the distributed perception agent re-evaluates the fault-sensitive fingerprint parameters every 5-10 minutes, dynamically adjusts and , drive the topology to evolve a new round, and ensure that it always adapts to the non-stationary working conditions of the system.

[0074] S411. Game equilibrium point extraction: When the number of connections that have not changed for three consecutive rounds in the topological stability index of S3 accounts for more than 80% of the total number of connections, the evolutionary game is determined to have reached a convergent state. The equilibrium point extracted at this time is the stable state of the connection strategy of the distributed sensing agent, which is represented by a fixed node connection matrix. To ensure the effectiveness of the equilibrium point, its stability needs to be verified by the policy volatility, that is, the change of the connection strategy within multiple consecutive periods needs to be controlled at an extremely low level.

[0075] The formula for calculating strategy volatility is:

[0076] ;

[0077] for The connection policy vector at time t, where each element of the connection policy vector is 0 or 1, representing disconnection or connection respectively; To verify the number of cycles, When this occurs, it indicates that the strategy volatility is less than [a certain value]. Confirm that the equilibrium point is valid;

[0078] S412. Equilibrium-Robust Response Mapping Model Construction: Taking the equilibrium point as the core input, a mapping model is constructed by combining the system robustness index. The model input includes the topological characteristics and fault-sensitive fingerprint of the equilibrium point, and the output is the robust response value that quantifies the robustness of the system. The model reflects the system's ability to cope with faults and disturbances under different equilibrium states.

[0079] Robust response value The calculation formula is:

[0080] ;

[0081] S413. Optimal network configuration solution and robustness threshold output: Through multi-disturbance simulation experiments, the impact of various typical disturbances on the system is simulated, the robust response values ​​under different equilibrium point topologies are tested, and the topology with higher robust response values ​​is selected as the optimal network configuration; the robustness threshold is calculated based on the optimal configuration, and this threshold is used to determine whether the system robustness meets the standard.

[0082] ;

[0083] The robustness threshold; The optimal robust response value; This is for safety margin.

[0084] Further, S5 includes:

[0085] S511. Game residual calculation, during system operation, the current connection strategy of the distributed perception agent needs to be monitored in real time, and compared with the game equilibrium point strategy extracted in S4, the deviation degree of the two is calculated, that is, the game residual, the residual is the core index to measure the deviation degree of the current state of the system from the stable state, the greater the residual, the farther the system deviates from the equilibrium state, which needs to be adjusted to return to stability;

[0086] Game residual The calculation formula is:

[0087] ;

[0088] The current connection strategy of the agent , the strategy at the equilibrium point , and the total number of agents ;

[0089] S512. Residual-delay mapping and parameter adjustment, based on the calculated game residual and residual delay time, the strategy drift rate and information compensation factor weight are dynamically adjusted through the residual-delay mapping function; when the residual is large and the duration is long, the strategy drift rate needs to be increased to promote the distributed perception agent to actively adjust the connection strategy, while the information compensation factor is reduced to enhance the sensitivity to deviation and accelerate the system to return to the stable state; otherwise, the adjustment amplitude is reduced to maintain the smooth operation of the system;

[0090] Strategy drift rate update formula:

[0091] Information compensation factor update formula: ;

[0092] Wherein, , is the initial value, is the adjustment coefficient, is the delay time of residual duration; , is the adjusted strategy drift rate and information compensation factor;

[0093] S513. According to the adjusted and , the perception topology structure and operation parameters are corrected in real time, the whole process does not need to retrain the model or offline modeling, realizes zero delay adaptation of non-stationary working conditions; topology correction includes expansion and contraction, and parameter correction involves communication bandwidth, sampling frequency, etc.

[0094] When , the topology expansion is triggered, and the number of connections is increased:

[0095] ;

[0096] When , trigger topology contraction, connection reduction number is:

[0097] .

[0098] Further, the method for real-time comparison of game convergence entropy and reusable robust threshold, triggering different levels of early warning according to the degree of over-threshold value is:

[0099] The game convergence entropy is compared with the robust threshold value in real time, when , triggering mild or severe early warning according to the degree of over-threshold value, mild early warning starting online updating mechanism of S5, severe early warning triggering system level reconstruction or maintenance process; after processing, re-executing the mapping optimization of S4 and the residual adjustment of S5, forming a perception-evaluation-warning-optimization closed loop, realizing the whole life cycle health management of energy storage system;

[0100] Among them, the game convergence entropy calculates the game convergence entropy, which is a comprehensive index to measure the convergence speed and stability of system strategy, according to the convergence speed and stability of distributed perception intelligent agent strategy; the greater the entropy value, the worse the system strategy convergence consistency, the lower the stability, and the higher the possibility of fault risk; otherwise, the system state is more stable;

[0101] The game convergence entropy is calculated by the following formula:

[0102] ;

[0103] ;

[0104] is the normalized probability of intelligent agent ; is the reciprocal of the time when intelligent agent reaches equilibrium; is the strategy fluctuation variance.

[0105] In summary, due to the adoption of the above technical solutions, the beneficial technical effects of the invention are:

[0106] Firstly, the unified representation and cross-domain association of heterogeneous signals are realized:

[0107] The application maps heterogeneous signals of electrochemistry, inertia and thermal energy storage to a unified high-dimensional state space based on a coupling function family based on the law of conservation of energy, and outputs inter-domain coupling gain factors, solving the defect that cross-domain signals are difficult to analyze in the prior art. The construction of the high-dimensional space makes the signals such as voltage, speed and temperature difference have consistent mathematical representation forms, and the inter-domain coupling gain factors quantify the mutual influence strength of different energy storage domains, providing a unified coordinate system for fault sensitive analysis, and significantly improving the recognition accuracy of multi-source collaborative fault.

[0108] In the second aspect, the dynamic adaptability and robustness of the network topology are improved.

[0109] The application introduces a strategy drift rate and an information compensation factor through the perception topology ad hoc network technology driven by evolutionary game, so that the network can dynamically adjust the connection strategy (expand the topology in the early stage of failure, and shrink the topology in normal working conditions) under the condition of information loss, solving the problem of poor adaptability of the fixed topology. At the same time, the game equilibrium-robust response mapping optimization and the residual error driven online updating mechanism ensure that the network still maintains high sensitivity under limited communication or non-stationary working conditions, and can adapt to scenes such as aging and load mutation without retraining the model, thereby reducing the operation and maintenance cost.

[0110] In the third aspect, a full life cycle health management closed loop is constructed.

[0111] The application quantifies the strategy convergence speed and stability as entropy values through the game convergence entropy health early warning closed loop, realizes multi-level early warning in combination with a robust threshold, triggers system reconstruction or maintenance process, and forms a complete closed loop of perception-evaluation-early warning-optimization. The mechanism solves the defect that the prior art lacks full process management, realizes full life cycle coverage from signal acquisition to fault early warning and system optimization, and significantly improves the operation safety and reliability of the multi-source energy storage system, prolonging the service life of the equipment. BRIEF DESCRIPTION OF DRAWINGS

[0112] Figure 1 The figure is a flowchart of a multi-source energy storage operation performance and fault state characterization parameter optimization method. DETAILED DESCRIPTION

[0113] In order to make the purpose, technical scheme and advantages of the invention more clear and explicit, the invention will be further described in detail in combination with embodiments. It should be understood that the specific embodiments described herein are only used to explain the invention and do not limit the invention.

[0114] As shown in Figure 1 The state perception network optimization method for multi-source energy storage operation performance and fault state characterization parameters comprises the following steps:

[0115] S1. Multi-domain signal synchronous acquisition and coupling mapping: acquire voltage signal, speed signal, and temperature difference signal, construct a family of coupling functions based on the law of energy conservation, map heterogeneous signals to a unified high-dimensional state space, and output inter-domain coupling gain factor to provide a consistent coordinate system for subsequent fault sensitivity analysis.

[0116] The electrochemical energy storage module collects voltage signals, the inertial energy storage unit collects rotational speed signals, and the thermal energy storage device collects temperature difference signals. The acquisition process uses BeiDou time synchronization to ensure time synchronization, with the synchronization error controlled within 10μs, ensuring the consistency of signals in different domains in the time dimension.

[0117] The step of constructing a family of coupled functions based on the law of energy conservation to map heterogeneous signals to a unified high-dimensional state space includes:

[0118] S111. Based on the law of conservation of energy, the conversion relationship between electrical energy in electrochemical energy storage, kinetic energy in inertial energy storage and thermal energy in thermal energy storage is analyzed. A family of coupled functions containing multiple sets of mapping relationships such as voltage-speed, speed-temperature difference, and voltage-temperature difference is constructed to realize the correlation quantization of energy signals in different domains.

[0119] The core expression of the family of coupling functions is the total energy coupling function, which takes the following form:

[0120] ;

[0121] This represents the value of the total energy coupling function; Electrical energy stored through electrochemical processes; Kinetic energy stored as inertia; Thermal energy for thermal energy storage; For the domain weight coefficients, satisfying ; This is the inter-domain coupling coefficient, used to quantify the interaction strength between energies in different domains;

[0122] S112. Heterogeneous signal mapping: The acquired heterogeneous signals (voltage, rotational speed, temperature difference) are converted into energy parameters of the corresponding domains. , , Then, through the above family of coupling functions, coordinate transformation is performed to map to a unified high-dimensional state space with dimensions of 10-20, thereby realizing the standardized expression of heterogeneous signals;

[0123] S113. Output Inter-domain Coupling Gain Factor: During the mapping of heterogeneous signals to a high-dimensional state space, the inter-domain coupling gain factor is calculated and output in real time. This factor quantifies the mutual influence intensity between signals from different domains, providing a consistent coordinate system and cross-domain correlation benchmark for subsequent fault sensitivity analysis. The formula for calculating the inter-domain coupling gain factor is:

[0124] ;

[0125] is the inter-domain coupling gain factor; respectively represent any two domains of electrochemistry, inertia, and thermal energy storage; The greater the value of the domain and the domain signal is, the stronger the coupling effect of the two domains.

[0126] S2. Asymmetric disturbance response spectrum construction, controllable asymmetric micro-perturbation is injected in a unified high-dimensional state space, a closed-loop disturbance-response experiment framework is constructed, response asymmetry, delay coupling coefficient and dissipation offset are extracted, and a unique fault sensitive fingerprint is formed after normalization processing;

[0127] The asymmetric micro-perturbation is defined as:

[0128] ;

[0129] is the micro-perturbation signal of the i-th dimension in the high-dimensional state space; 、 are the amplitudes of the forward and reverse micro-perturbations, respectively, ; 、 are the action times of the forward and reverse micro-perturbations, respectively, ; is the micro-perturbation angular frequency, which maintains a certain proportion (ωi=ω0 / i) of the system inherent frequency to excite significant response;

[0130] The amplitude of the controllable asymmetric micro-perturbation is 0.1%-5% of the normal signal amplitude, and has asymmetry in the positive and negative directions and action time;

[0131] The experimental framework includes a micro-perturbation generator, a high-dimensional state monitoring module, a response signal acquisition module, and a feedback control unit. The experimental framework ensures that the micro-perturbation is injected according to the preset strategy, and the response signal of the system under the action of the micro-perturbation is acquired in real time and synchronously:

[0132] Among them, the response asymmetry quantifies the response difference of the system to the forward and reverse asymmetric micro-perturbations. The greater the difference, the more significant the fault precursor; the calculation formula is:

[0133] ;

[0134] denotes the response asymmetry, and the greater the value, the more significant the response difference of the system to the asymmetric micro-perturbation;

[0135] is the dimension of the high-dimensional space;​ This represents the peak value of the positive perturbation response; This represents the peak value of the reverse perturbation response; Peak value of the reverse perturbation response Corresponding response time; Peak value of positive perturbation response Corresponding response time;

[0136] The delay coupling coefficient characterizes the time delay and correlation strength between the perturbation signal and the response signal, reflecting anomalies in the system's dynamic characteristics; the calculation formula is:

[0137] ;

[0138] The delay coupling coefficient; It is a perturbation signal vector containing perturbation values ​​for each dimension of the high-dimensional space;

[0139] The response signal vector contains response values ​​for each dimension of the high-dimensional space; For delay time; , These are the mean vectors of the perturbation and the response, respectively, used to eliminate the influence of the DC component;

[0140] The dissipation offset measures the deviation of a system's energy dissipation under perturbation from its normal state, reflecting anomalies in energy conversion efficiency. The calculation formula is as follows:

[0141] ;

[0142] For dissipation offset, A higher value indicates a more significant abnormality in system energy dissipation (such as when the thermal insulation layer fails). It will increase significantly). This represents the total duration of the perturbation. The L2 norm characterizes the energy of the perturbation input. The L2 norm of the response signal characterizes the energy of the system response.

[0143] The fault-sensitive fingerprint, as a core feature distinguishing different fault types (such as battery aging, flywheel imbalance, and thermal storage pipe blockage), is used to quantify the sensitivity differences of different fault precursors. The fault-sensitive fingerprint is represented as follows:

[0144] ;

[0145] ; ; ;

[0146] in, is the normalized response asymmetry; is the normalized delay coupling coefficient; is the normalized dissipation offset.

[0147] S3. Evolutionary game driven perception topology self-organizing network, taking fault sensitive fingerprint as input, deploying distributed perception agent at key nodes of energy storage system; introducing strategy drift rate and information compensation factor as game variables, evolving connection strategy under information missing condition, dynamically generating elastic topology;

[0148] The distributed perception agent has the ability to collect local state signals (such as battery voltage, flywheel vibration frequency, heat storage medium flow rate) of the node it is located in, process local data based on built-in algorithms (such as calculating local fault probability), and exchange information with adjacent distributed perception agents through wireless or wired means (communication distance is usually 5-10 meters, ensuring low delay); At the same time, the initial calculation resources (such as 32-bit microprocessor, operation capacity ≥100MIPS), communication bandwidth (≥1Mbps) and storage capacity (≥1MB) are allocated to the distributed perception agent to ensure that it can carry out game calculation and information exchange tasks.

[0149] The strategy drift rate represents the probability of the distributed perception agent changing the current connection strategy, reflecting the flexibility of topology adjustment. Its value is related to the fault sensitive fingerprint. When the fault sensitive fingerprint parameter fluctuates greatly (such as >0.5, that is, the dissipation offset is significant), the drift rate is in the range of 0.3-0.5, promoting the distributed perception agent to actively adjust the connection; When the fault sensitive fingerprint is stable (such as <0.2), the drift rate is reduced to 0.1-0.2, maintaining the stability of the topology;

[0150] The information compensation factor is used to make up for the information missing caused by communication delay or signal loss, and its value is in the range of 0.5-1.0; When the information missing degree (the proportion of missing fingerprint parameters) is >30%, the factor value is increased to 0.8-1.0, and the missing data is compensated through information sharing of adjacent distributed perception agents; When the information integrity is ≥90%, the factor is reduced to 0.5-0.7, reducing redundant calculation.

[0151] Evolution of connection strategy under information missing condition, dynamic generation of elastic topology includes the following steps:

[0152] S311. Game rule initialization and benefit function construction:

[0153] Rule setting, the core game rule of the distributed sensing agent is to maximize the information benefit and minimize the communication cost, and the strategy evaluation is performed every 50-100 ms; the evaluation period is dynamically adjusted according to the fluctuation degree of the fault sensitive fingerprint: when (normalized dissipation offset) > 0.5, it is shortened to 50 ms to quickly respond to faults; when (normalized delay coupling coefficient) < 0.2, it is extended to 100 ms to reduce resource consumption.

[0154] Benefit function construction, based on the functional characteristics of the distributed sensing agent, the benefit function is constructed as the basis for strategy adjustment: ; wherein, is a weight coefficient, increases with , and ; the information integrity is the proportion of effective fingerprint parameters obtained by the distributed sensing agent, for example, when only one fingerprint parameter is missing, the integrity = 2 / 3 ≈ 0.67; the communication cost is positively correlated with the number of connected nodes (the cost increases by 0.1 for each additional connection); is an information compensation factor, the missing information estimate value is derived from the shared data of adjacent agents (such as estimating the of the battery management unit and the of the flywheel control module to estimate the of the thermal storage system).

[0155] S312. Connection strategy evolution:

[0156] Strategy adjustment trigger, according to the current strategy drift rate and the information compensation factor , the distributed sensing agent decides whether to adjust the connection strategy:

[0157] When the current strategy drift rate is 0.3-0.5 (large fingerprint fluctuation), there is a 30%-50% probability of triggering strategy adjustment every round of evaluation; when the current strategy drift rate is 0.1-0.2 (stable fingerprint), the adjustment probability is reduced to 10%-20%, reducing unnecessary topology changes;

[0158] Information missing processing, if the information missing degree > 30%, the distributed sensing agent preferentially sends information request to adjacent nodes, and the calculation formula of the information sharing weight is: ;

[0159] The higher the value of the information sharing weight of a node, the higher the priority of receiving information sharing request, ensuring that the missing data is efficiently compensated;

[0160] S313. Elastic topology expansion, when the benefit function (Information benefit is significantly higher than communication cost) triggers topology expansion:

[0161] Expansion logic, distributed sensing agents increase connections with high information value nodes,

[0162] Expansion quantity calculation formula:

[0163] ;

[0164] Where, is the expansion coefficient, is the rounding function; is the current strategy drift rate;

[0165] Expansion target, preferentially connect nodes with strong complementary fault sensitivity fingerprints (such as battery management unit agents preferentially connecting temperature regulation node agents to obtain normalized dissipation offset information), expand the topology diameter (the furthest node communication hop count) from the initial 3-4 hops to 5-6 hops, covering a wider fault sensing range;

[0166] S314. Elastic topology contraction, when the benefit function U < 0.3 (communication cost is too high or information benefit is limited), triggers topology contraction:

[0167] Distributed sensing agents disconnect connections with low information value nodes, and the contraction quantity

[0168] Calculation formula: ;

[0169] Where, is the contraction coefficient;

[0170] Contraction target, preferentially disconnect connections with information repetition rate > 80% (such as two adjacent battery management unit distributed sensing agents with high information overlap), contract the topology diameter to the initial level, reduce communication energy consumption (after contraction, energy consumption can be reduced by 20%-30%);

[0171] S315. Topology stability verification and iteration:

[0172] Stability verification, after each expansion or contraction, calculate the connection stability index of the topology:

[0173] ;

[0174] When ≥ 0.8, it is determined that the topology enters a stable state; if ​If < 0.8, repeat steps S312-S314 until the topology converges.

[0175] Iterative update, as the system state changes (such as fault mitigation or aggravation), the distributed sensing agent re-evaluates the fault sensitive fingerprint parameters every 5-10 minutes, dynamically adjusts and , drive the topology to a new round of evolution, to ensure that it is always adapted to the non-stationary working conditions of the system.

[0176] S4. Game equilibrium-robust response mapping optimization, from the evolutionary game process of the elastic topology, extract the equilibrium point after convergence, establish the equilibrium-robust response mapping model, solve the optimal network configuration through multi-perturbation simulation, and output the reusable robust threshold, which provides a benchmark for state evaluation;

[0177] S411. Game equilibrium point extraction, when the number of connections that do not change continuously for 3 rounds in S3 is more than 80% of the total number of connections, it is determined that the evolutionary game has reached a convergent state. The equilibrium point extracted at this time is the stable state of the distributed sensing agent connection strategy, which is represented by a fixed node connection matrix. For example, the battery management unit agent and the temperature regulation node distributed sensing agent form a stable connection, with a connection weight ≥ 0.7 and no fluctuation within 500ms. To ensure the effectiveness of the equilibrium point, it needs to be verified for stability by the strategy volatility rate, that is, the variation amplitude of the connection strategy in consecutive multiple periods needs to be controlled at a very low level.

[0178] The calculation formula of the strategy volatility rate is:

[0179] ;

[0180] For the connection strategy vector at the moment, the connection strategy vector element is 0 or 1, indicating disconnection or connection respectively. is the verification period (usually 10), , it means that the strategy fluctuation is less than confirm the effectiveness of the equilibrium point;

[0181] S412. Construction of equilibrium-robust response mapping model, taking the equilibrium point as the core input, combining the system robustness index (fault recognition accuracy, anti-interference ability, response speed) to construct the mapping model, the model input includes the topology characteristics (connection density, average path length) of the equilibrium point and the fault sensitive fingerprint, and the output is the robust response value of the quantitative system robustness. The model reflects the system's ability to respond to faults and disturbances under different equilibrium states.

[0182] The calculation formula of the robust response value is:

[0183] ;

[0184] S413. Optimal network configuration solution and robust threshold output: Through multi-disturbance simulation experiments, the impact of various typical disturbances (such as sudden voltage rise, sudden speed drop, etc.) on the system is simulated, and the robust response values ​​under different equilibrium point topologies are tested. The topology with higher robust response values ​​is selected as the optimal network configuration. Based on the optimal configuration, the robust threshold is calculated, which is used to determine whether the system robustness meets the standard.

[0185] ;

[0186] The robustness threshold; The optimal robust response value; This is for a safety margin.

[0187] S5. The residual-driven online update mechanism calculates the deviation between the current connection strategy and the game equilibrium point of the distributed sensing agent in real time, and dynamically adjusts the weights of the strategy drift rate and information compensation factor using the residual-delay mapping function to achieve zero-delay correction of the sensing topology and operating parameters, adapting to non-stationary working conditions.

[0188] S511. Game Residual Calculation: During system operation, it is necessary to monitor the current connection strategy of the distributed sensing agent in real time and compare it with the game equilibrium point strategy extracted in S4. The degree of deviation between the two is calculated, which is the game residual. The residual is the core indicator for measuring the degree of deviation between the current state and the stable state of the system. The larger the residual, the further the system deviates from the equilibrium state, and it is necessary to adjust the strategy to return to stability.

[0189] Game residuals The calculation formula is:

[0190] ;

[0191] For intelligent agents The current connection policy, The strategy at the equilibrium point, The total number of agents;

[0192] S512. Residual-Delay Mapping and Parameter Adjustment: Based on the calculated game residuals and the delay time of the residuals, the weights of the policy drift rate and the information compensation factor are dynamically adjusted through the residual-delay mapping function. When the residuals are large and the duration is long, the policy drift rate needs to be increased to encourage the distributed sensing agent to actively adjust the connection strategy, while the information compensation factor is reduced to enhance the sensitivity to deviations and accelerate the system's return to a stable state. Conversely, the adjustment magnitude is reduced to maintain the smooth operation of the system.

[0193] Strategy drift rate update formula:

[0194] Information compensation factor update formula:

[0195] wherein, is the initial value, is the adjustment coefficient, is the delay time of residual persistence; is the adjusted strategy drift rate and information compensation factor;

[0196] S513. According to the adjusted and , the perception topology and operating parameters are corrected in real time, and the whole process does not need to retrain the model or offline modeling, realizing zero delay adaptation of non-stationary working conditions (such as voltage fluctuation caused by battery aging and flywheel speed sudden change caused by load mutation). Topology correction includes expansion (increasing connections) and contraction (reducing connections), and parameter correction involves communication bandwidth, sampling frequency, etc.

[0197] When (moderate deviation or above), topology expansion is triggered, and the number of connections to be increased is:

[0198]

[0199] When (minimal deviation), topology contraction is triggered, and the number of connections to be reduced is:

[0200]

[0201] S6. Game convergence entropy health warning closed loop, according to the convergence speed and stability of the distributed perception intelligent agent, the game convergence entropy is calculated, the game convergence entropy is compared with the reusable robustness threshold in real time, and different levels of warning are triggered according to the degree of exceeding the threshold.

[0202] The steps of comparing the game convergence entropy with the reusable robustness threshold in real time and triggering different levels of warning according to the degree of exceeding the threshold include:

[0203] When the calculated game convergence entropy is compared with the robustness threshold output by S4 in real time, when , mild or severe warning is triggered according to the degree of exceeding the threshold, the online update mechanism of S5 is started, and severe warning triggers system-level reconstruction or maintenance process, and after processing, the mapping optimization of S4 and the residual adjustment of S5 are executed again, forming a perception-evaluation-warning-optimization closed loop, realizing the whole life cycle health management of the energy storage system.​​​​​

[0204] The game convergence entropy calculation is a comprehensive indicator that measures the convergence speed and stability of a system strategy. It is calculated based on the convergence speed (the reciprocal of the time to reach the equilibrium point) and stability (variance of strategy fluctuation) of the agent's strategy. The larger the entropy value, the worse the convergence consistency of the system strategy, the lower the stability, and the higher the possibility of failure risk; conversely, the more stable the system state is.

[0205] Game convergence entropy The calculation formula is:

[0206] ;

[0207] ;

[0208] For intelligent agents The normalized probability; intelligent agent The reciprocal of the time required to reach equilibrium; Strategy volatility variance.

[0209] The above description is a preferred embodiment of the invention and is not intended to limit the scope of the invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A method for optimizing the operating performance and fault state characterization parameters of multi-source energy storage, characterized in that, Includes the following steps: S1. Multi-domain signal synchronous acquisition and coupling mapping: acquire voltage signal, speed signal, and temperature difference signal; construct a family of coupling functions based on the law of energy conservation; map heterogeneous signals to a unified high-dimensional state space; and output inter-domain coupling gain factor. S2. Construction of asymmetric perturbation response spectrum: controllable asymmetric perturbation is injected into a unified high-dimensional state space to construct a closed-loop perturbation-response experimental framework. The response asymmetry, delay coupling coefficient and dissipation offset are extracted and normalized to form a unique fault-sensitive fingerprint. S3. Evolutionary game-driven sensing topology self-organizing network: Using a unique fault-sensitive fingerprint as input, distributed sensing agents are deployed at key nodes of the energy storage system. The policy drift rate and information compensation factor are introduced as game variables. The connection strategy evolves under the condition of missing information, and the elastic topology is dynamically generated. S4. Game Equilibrium-Robust Response Mapping Optimization: Extract the converged equilibrium point from the evolutionary game process of elastic topology, establish an equilibrium-robust response mapping model, solve the optimal network configuration through multi-perturbation simulation, and output a reusable robust threshold. S5. The residual-driven online update mechanism calculates the deviation between the current connection strategy and the game equilibrium point of the distributed sensing agent in real time, and dynamically adjusts the weights of the strategy drift rate and information compensation factor using the residual-delay mapping function to achieve zero-delay correction of the sensing topology and operating parameters. S6. Game convergence entropy health early warning closed loop: Based on the convergence speed and stability of the distributed sensing agent, calculate the game convergence entropy, compare the game convergence entropy with the reusable robust threshold in real time, and trigger different levels of early warning according to the degree of exceeding the threshold.

2. The method for optimizing the operating performance and fault state characterization parameters of multi-source energy storage according to claim 1, characterized in that, The voltage signal is acquired through an electrochemical energy storage module, the rotation speed signal is acquired through an inertial energy storage unit, and the temperature difference signal is acquired through a thermal energy storage device. The acquisition process uses BeiDou time synchronization to ensure time synchronization, with the synchronization error controlled within 10μs, ensuring the consistency of signals in different domains in the time dimension.

3. The method for optimizing the operating performance and fault state characterization parameters of multi-source energy storage according to claim 1, characterized in that, The steps of constructing a family of coupling functions based on the law of energy conservation, mapping heterogeneous signals to a unified high-dimensional state space, and outputting an inter-domain coupling gain factor include: S111. Construct a family of coupled functions. Based on the law of conservation of energy, analyze the conversion relationship between electrical energy in electrochemical energy storage, kinetic energy in inertial energy storage, and thermal energy in thermal energy storage. Construct a family of coupled functions that includes multiple sets of mapping relationships such as voltage-speed, speed-temperature difference, and voltage-temperature difference to achieve correlation quantization of energy signals in different domains. The core expression of the family of coupling functions is the total energy coupling function, which takes the following form: ； This represents the value of the total energy coupling function; Electrical energy stored through electrochemical processes; Kinetic energy stored as inertia; Thermal energy for thermal energy storage; For the domain weight coefficients, satisfying ; This is the inter-domain coupling coefficient, used to quantify the interaction strength between energies in different domains; S112. Heterogeneous signal mapping: The acquired heterogeneous signals are converted into energy parameters of the corresponding domains, and then the coordinate transformation is performed through the coupling function family to map them to a unified high-dimensional state space with a dimension of 10-20, thereby realizing the standardized expression of heterogeneous signals. S113. Output inter-domain coupling gain factor: During the mapping of heterogeneous signals to a high-dimensional state space, the inter-domain coupling gain factor is calculated and output in real time; the formula for calculating the inter-domain coupling gain factor is: ； This is the inter-domain coupling gain factor; They represent any two domains in electrochemistry, inertia, and thermal energy storage, respectively. The larger the value, the more likely it is that the first... Domain and First The stronger the coupling effect of the domain signal.

4. The method for optimizing the operating performance and fault state characterization parameters of multi-source energy storage according to claim 1, characterized in that, The asymmetric perturbation is defined as: ； It is the perturbation signal of the th dimension in the high-dimensional state space; , These represent the amplitudes of the positive and negative perturbations, respectively. ; , These represent the duration of the positive and negative perturbations, respectively. ; The perturbation angular frequency is kept in a certain proportion to the system's natural frequency in order to elicit a significant response; The amplitude of the controllable asymmetric perturbation is 0.1%-5% of the normal signal amplitude, and it is asymmetric in the positive and negative directions and in the duration of action.

5. The method for optimizing the operating performance and fault state characterization parameters of multi-source energy storage according to claim 1, characterized in that, The experimental framework includes a perturbation generator, a high-dimensional state monitoring module, a response signal acquisition module, and a feedback control unit. The experimental framework ensures that perturbations are injected according to a preset strategy and acquires the system's response signals under the action of perturbations in real time and synchronously. Among them, response asymmetry measures the difference in the system's response to forward and reverse asymmetric perturbations; the greater the difference, the more significant the precursor to a fault; the calculation formula is: ； This indicates the degree of response asymmetry; a larger value indicates a more significant difference in the system's response to asymmetric perturbations. It is a high-dimensional space dimension; This represents the peak value of the positive perturbation response; This represents the peak value of the reverse perturbation response; Peak value of the reverse perturbation response Corresponding response time; Peak value of positive perturbation response Corresponding response time; The delay coupling coefficient characterizes the time delay and correlation strength between the perturbation signal and the response signal, reflecting anomalies in the system's dynamic characteristics; the calculation formula is: ； The delay coupling coefficient; It is a perturbation signal vector containing perturbation values ​​for each dimension of the high-dimensional space; The response signal vector contains response values ​​for each dimension of the high-dimensional space; For delay time; , These are the mean vectors of the perturbation and the response, respectively, used to eliminate the influence of the DC component; The dissipation offset measures the deviation of a system's energy dissipation under perturbation from its normal state, reflecting anomalies in energy conversion efficiency. The calculation formula is as follows: ； For dissipation offset, The larger the value, the more obvious the abnormal energy dissipation of the system; This represents the total duration of the perturbation. The L2 norm characterizes the energy of the perturbation input. The L2 norm of the response signal characterizes the energy of the system response.

6. The method for optimizing the operating performance and fault state characterization parameters of multi-source energy storage according to claim 1, characterized in that, The fault-sensitive fingerprint, as a core feature distinguishing different fault types, is used to quantify the sensitivity differences of different fault precursors. The fault-sensitive fingerprint is represented as follows: ； ； ； ； in, The normalized response asymmetry; The normalized delay coupling coefficient; This is the normalized dissipation offset.

7. The method for optimizing the operating performance and fault state characterization parameters of multi-source energy storage according to claim 1, characterized in that, The distributed sensing agent has the ability to collect local state signals of its node in real time, perform local data processing based on built-in algorithms, and interact with neighboring agents wirelessly or via wired means. At the same time, the distributed sensing agent is allocated initial computing resources, communication bandwidth and storage capacity to ensure that it can carry out game calculation and information interaction tasks. The policy drift rate represents the probability that an agent will change its current connection policy, reflecting the flexibility of topology adjustment. The policy drift rate is related to the fault-sensitive fingerprint. When the fault-sensitive fingerprint parameters fluctuate greatly, the drift rate ranges from 0.3 to 0.5, which promotes the distributed sensing agent to actively adjust the connection. When the fault-sensitive fingerprint is stable, the drift rate drops to 0.1-0.2, maintaining topology stability. The information compensation factor is used to compensate for information loss caused by communication delay or signal loss, and its value ranges from 0.5 to 1.

0. When the information loss is greater than 30%, the factor value is increased to 0.8-1.0 to compensate for missing data through information sharing between neighboring distributed sensing agents. When the information integrity is greater than or equal to 90%, the factor is reduced to 0.5-0.7 to reduce redundant calculations.

8. The method for optimizing the operating performance and fault state characterization parameters of multi-source energy storage according to claim 1, characterized in that, The process of introducing strategy drift rate and information compensation factor as game variables to evolve connection strategies and dynamically generate elastic topologies under conditions of information deficiency includes the following steps: S311. Game rule initialization and payoff function construction: The rules are set so that the distributed sensing agent uses maximizing information gain and minimizing communication cost as the core game rule, and performs a policy evaluation every 50-100ms; the evaluation cycle is dynamically adjusted according to the fluctuation of the fault-sensitive fingerprint: when When >0.5, shorten to 50ms for rapid fault response; when When <0.2, extend to 100ms to reduce resource consumption; Revenue function construction: Based on the functional characteristics of distributed sensing agents, a revenue function is constructed. As a basis for determining strategy adjustments: ; in, These are the weighting coefficients. Follow Increase and improve, and Information completeness is the percentage of valid fingerprint parameters acquired by the distributed sensing agent; communication cost is positively correlated with the number of connected nodes. As an information compensation factor, the estimated value of missing information is derived from the shared data of neighboring distributed sensing agents; S312. Evolution of Connection Strategy: Strategy adjustment is triggered based on the current strategy drift rate. and information compensation factor Whether the distributed sensing agent should adjust the connection strategy: When the current strategy drift rate is between 0.3 and 0.5, there is a 30% to 50% probability of triggering strategy adjustment in each round of evaluation; when the current strategy drift rate is between 0.1 and 0.2, the adjustment probability drops to 10% to 20%, reducing unnecessary topology changes. For handling missing information, if the missing information rate is greater than 30%, the distributed sensing agent prioritizes sending information requests to neighboring nodes, and the information sharing weight is adjusted accordingly. The calculation formula is: ; Information sharing weight Nodes with higher values ​​receive information sharing requests first, ensuring that missing data is efficiently compensated. S313. Elastic topology extension, when the payoff function At that time, topology expansion is triggered: Extending the logic, the distributed sensing agent adds connections with nodes of high information value. Formula for calculating the number of extensions: ; in, For expansion coefficient, It is a rounding function; This represents the current strategy drift rate; Expand the target by prioritizing the connection of nodes with strong complementarity in fault-sensitive fingerprints, thereby expanding the topology diameter from the initial 3-4 hops to 5-6 hops and covering a wider range of fault perception. S314. Elastic topology contraction: Topology contraction is triggered when the payoff function U < 0.

3. Distributed sensing agents disconnect from nodes with low information value and reduce their number of nodes. Calculation formula: ; in, The shrinkage coefficient; By shrinking the target, connections with information redundancy rates >80% are prioritized for disconnection, thereby shrinking the topology diameter to the initial level and reducing communication energy consumption. S315. Topological stability verification and iteration: Stability verification involves calculating the connectivity stability index of the topology after each expansion or contraction. : ； when When the value is ≥0.8, the topology is considered to have entered a stable state; if If <0.8, repeat steps S312-S314 until the topology converges; Iterative updates: As the system state changes, the distributed sensing agent re-evaluates the fault-sensitive fingerprint parameters every 5-10 minutes and dynamically adjusts them. and This drives the topology to undergo a new round of evolution, ensuring that it always adapts to the non-stationary operating conditions of the system.

9. The method for optimizing the operating performance and fault state characterization parameters of multi-source energy storage according to claim 1, characterized in that, Step S4 includes: S411. Game equilibrium point extraction: When the number of connections that have not changed for three consecutive rounds in the topological stability index in step S3 accounts for more than 80% of the total number of connections, the evolutionary game is determined to have reached a convergent state. The equilibrium point extracted at this time is the stable state of the connection strategy of the distributed sensing agent, which is represented by a fixed node connection matrix. To ensure the effectiveness of the equilibrium point, its stability needs to be verified by the policy volatility, that is, the change of the connection strategy within multiple consecutive periods needs to be controlled at an extremely low level. The formula for calculating strategy volatility is: ； for The connection policy vector at time t, where each element of the connection policy vector is 0 or 1, representing disconnection or connection respectively; To verify the number of cycles, When this occurs, it indicates that the strategy volatility is less than [a certain value]. Confirm that the equilibrium point is valid; S412. Equilibrium-Robust Response Mapping Model Construction: Taking the equilibrium point as the core input, a mapping model is constructed by combining the system robustness index. The model input includes the topological characteristics and fault-sensitive fingerprint of the equilibrium point, and the output is the robust response value that quantifies the robustness of the system. The model reflects the system's ability to cope with faults and disturbances under different equilibrium states. Robust response value The calculation formula is: ； S413. Optimal network configuration solution and robustness threshold output: Through multi-disturbance simulation experiments, the impact of various typical disturbances on the system is simulated, the robust response values ​​under different equilibrium point topologies are tested, and the topology with higher robust response values ​​is selected as the optimal network configuration; the robustness threshold is calculated based on the optimal configuration, and this threshold is used to determine whether the system robustness meets the standard. ； The robustness threshold; The optimal robust response value; This is for safety margin.

10. The method for optimizing the operating performance and fault state characterization parameters of multi-source energy storage according to claim 1, characterized in that, Step S5 includes: S511. Game residual calculation: During system operation, it is necessary to monitor the current connection strategy of the distributed sensing agent in real time and compare it with the game equilibrium point strategy extracted in step S4. The degree of deviation between the two is calculated, i.e., the game residual. The residual is the core indicator for measuring the degree of deviation between the current state and the stable state of the system. The larger the residual, the further the system deviates from the equilibrium state, and it is necessary to adjust the strategy to return to stability. Game residuals The calculation formula is: ； For intelligent agents The current connection policy, The strategy at the equilibrium point, The total number of agents; S512. Residual-Delay Mapping and Parameter Adjustment: Based on the calculated game residuals and the delay time of the residuals, the weights of the policy drift rate and the information compensation factor are dynamically adjusted through the residual-delay mapping function. When the residuals are large and the duration is long, the policy drift rate needs to be increased to encourage the agent to actively adjust the connection policy, while the information compensation factor is reduced to enhance the sensitivity to deviation and accelerate the system's return to a stable state. Conversely, the adjustment magnitude is reduced to maintain the smooth operation of the system. Policy drift rate update formula: Information compensation factor update formula: ; in, , As the initial value, To adjust the coefficient, The duration of the residual; , The adjusted strategy drift rate and information compensation factor; S513. According to the adjusted and It can correct the sensing topology and operating parameters in real time, without retraining the model or offline modeling, and achieve zero-latency adaptation to non-stationary conditions. Topology correction includes two methods: expansion and contraction, while parameter correction involves communication bandwidth and sampling frequency. when When this occurs, topology expansion is triggered, and the number of connections increases by: ； when When topology shrinkage is triggered, the number of connections is reduced by: 。 11. The method for optimizing the operating performance and fault state characterization parameters of multi-source energy storage according to claim 1, characterized in that, The method of comparing the game convergence entropy with a reusable robust threshold in real time and triggering different levels of early warning based on the degree of threshold exceedance is as follows: Game convergence entropy The robust threshold output from step S4 When performing real-time comparison, when If the threshold is exceeded, a mild or severe warning is triggered. A mild warning initiates the online update mechanism in step S5, while a severe warning triggers a system-level refactoring or maintenance process. After processing, the mapping optimization in step S4 and the residual adjustment in step S5 are re-executed to form a closed loop of perception-assessment-warning-optimization, thereby achieving full life-cycle health management of the energy storage system. The game convergence entropy calculation is a comprehensive indicator that measures the convergence speed and stability of the system strategy. It is calculated based on the convergence speed and stability of the distributed sensing agent strategy. The larger the entropy value, the worse the convergence consistency of the system strategy, the lower the stability, and the higher the possibility of failure risk. Conversely, the lower the entropy value, the more stable the system state. Game convergence entropy The calculation formula is: ； ； For intelligent agents The normalized probability; intelligent agent The reciprocal of the time required to reach equilibrium; Strategy volatility variance.

Images