Fault prediction method and system for source network load storage cooperative system based on digital twinning

Abstract

The invention relates to a source network load storage cooperative system fault prediction method and system based on digital twinning. The obstacle prediction system comprises a physical perception layer, a data interaction layer and a digital twin service layer. The fault prediction method comprises the following steps: data cleaning and preprocessing; the digital twinning state is synchronized; generating a residual error of the whole system, and calculating the deviation and the residual error of the key observed quantity; calculating a self-adaptive dynamic threshold value; classifying and positioning fault modes; a predictive maintenance suggestion is generated, and the system not only outputs a fault alarm, but also can predict the residual life according to a parameter identification result; compared with the prior art, the invention has the following beneficial effects: the anti-interference capability is extremely high; full-life-cycle monitoring is carried out; massive fault samples are not needed; and the fault positioning is accurate.

Description

Technical Field

[0001] This invention relates to a data processing system or method specifically applicable to power systems, and more specifically to a fault prediction method and system for a source-grid-load-storage coordinated system based on digital twins. Background Technology

[0002] With the acceleration of the global energy transition, traditional unidirectional power systems are gradually evolving into new power systems characterized by deep interaction between power generation, grid, load, and storage. These systems integrate intermittent renewable energy sources such as photovoltaic and wind power, electronic grid interfaces, flexible loads with demand response capabilities, and electrochemical energy storage devices. While this highly coupled system architecture improves energy efficiency and system flexibility, it also brings unprecedented operational risks and challenges in fault diagnosis.

[0003] In a power generation, grid, load, and energy storage system, the dynamic characteristics of each subsystem are intertwined. For example, fluctuations in photovoltaic output on the power supply side can cause voltage flicker on the grid side, leading to frequent charging and discharging switching on the energy storage side and even affecting the power quality on the load side. Traditional fault detection mechanisms based on single-device threshold protection, such as overvoltage and overcurrent protection, often only activate after a fault occurs and causes damage, lacking early warning capabilities.

[0004] Currently, fault prediction technologies for power generation, grid, load, and storage systems are mainly divided into three categories, but all of them have significant shortcomings in practical applications:

[0005] Signal processing-based methods utilize tools such as Fourier transform (FFT) and wavelet transform to analyze the spectral characteristics of voltage and current signals. These methods perform well under steady-state conditions, but the power generation, grid, load, and energy storage system exhibits strong non-stationarity and randomness. For example, cloud cover can cause sudden changes in photovoltaic output, and these transient signal variations can easily be misinterpreted as fault characteristics, leading to a high false alarm rate.

[0006] Data-driven artificial intelligence approaches utilize neural networks to build fault classification models. However, training high-precision AI models requires massive amounts of labeled fault data. In real-world industrial scenarios, fault data for critical equipment such as energy storage batteries is extremely scarce, and system operating conditions vary greatly. This results in AI models exhibiting poor generalization ability when faced with unseen fault modes and lacking interpretability, making it difficult for maintenance personnel to trust their tripping recommendations.

[0007] Traditional analytical modeling methods, such as Kalman filtering, rely on mechanistic models for state observation. These methods depend on precise mathematical models, but parameters in real-world systems, such as battery internal resistance and line impedance, drift over time. Models with fixed parameters will generate increasingly larger prediction errors as operating time increases, ultimately leading to monitoring failure.

[0008] Digital twin technology offers a new approach to fault prediction by constructing virtual mirrors of physical entities and using real-time data to drive the synchronous operation of virtual models. However, most existing digital twin fault diagnosis solutions are at the static monitoring stage, comparing only the deviation between real-time and theoretical values. In power-grid-load-storage systems, due to extremely rapid system condition switching, such as microgrid off-grid / grid-connected switching or high-power charging of electric vehicles, the derivatives of system state variables are extremely large. During these dynamic processes, due to communication delays, asynchronous sensor sampling, or differences in model dynamic response, large instantaneous residuals inevitably arise between real-time and twin values. If traditional fixed threshold judgment logic is used, these normal dynamic residuals will be falsely reported as system faults.

[0009] Therefore, there is an urgent need in this field for a digital twin fault prediction scheme with adaptive parameter updates that can adapt to the high dynamic characteristics of source-grid-load-storage systems, in order to solve the problems of high false alarm rate in dynamic processes and high missed alarm rate for minor faults. Summary of the Invention

[0010] This invention provides a fault prediction method and system for a source-grid-load-storage system based on digital twins. It aims to solve the problems of false alarms caused by drastic fluctuations in operating conditions when dealing with multi-source heterogeneous energy systems, as well as the difficulty in extracting early, subtle fault features. By constructing a real-time dynamic simulation model, i.e., a digital twin, encompassing the physical mechanisms of the source, grid, load, and storage, and introducing an adaptive residual analysis mechanism based on the system's global power dynamics, accurate assessment of equipment health status and fault trend prediction are achieved. The specific technical solution is as follows:

[0011] A fault prediction system for a source-grid-load-storage coordinated system based on digital twins includes: a physical sensing layer, a data interaction layer, and a digital twin service layer.

[0012] The physical sensing layer is the hardware entity of the source-grid-load-storage system, including the following subsystems and key sensor configurations:

[0013] A fault prediction system for a source-grid-load-storage coordinated system based on digital twins includes: a physical sensing layer, a data interaction layer, and a digital twin service layer.

[0014] The physical sensing layer is the hardware entity of the source-grid-load-storage system, including the following subsystems and key sensor configurations:

[0015] The source side, i.e., the photovoltaic power generation unit, includes: a high-precision Hall current sensor and a voltage divider resistor network deployed at the combiner box for a distributed photovoltaic array; a total radiation meter deployed on the array plane; and a meter for collecting backsheet temperature. Surface mount sensor;

[0016] The energy storage side, i.e., the lithium battery energy storage unit, includes: a lithium iron phosphate battery pack equipped with a BMS; and it collects the voltage of individual battery cells. The second voltage sensor collects the total current of the battery cluster. The second current sensor collects the temperature at key points. Temperature sensor; BMS broadcasts data externally via CAN bus.

[0017] The grid side, i.e., power conversion and interface, includes: a bidirectional DC / AC inverter connecting the DC bus and the AC grid, and a device for measuring the three-phase voltage on the AC side. The third voltage sensor measures the three-phase current on the AC side. The third current sensor measures the DC bus voltage. The fourth voltage sensor.

[0018] The load side, or flexible load, includes: controllable lighting loads, electric vehicle charging stations; and real-time power data from smart meters. Power sensor, power factor acquisition The power factor acquisition device.

[0019] The data interaction layer includes: an industrial edge computing gateway based on ARM architecture, which collects data from each subsystem through bus polling; a communication protocol, which uses the MQTT protocol to upload the cleaned data to the digital twin server; and an NTP or PTP protocol to ensure that the data from the source, network, load, and storage sides have a unified timestamp.

[0020] The digital twin service layer is deployed on a high-performance server or cloud platform and runs the digital twin model; it includes: a numerical integral solver written in Python or C++ for real-time solving of the differential equations described below; and a fault analyzer that executes adaptive threshold calculation and fault logic judgment algorithms.

[0021] Furthermore, the digital twin model uses the state-space method to establish a unified system model; the photovoltaic power generation unit uses a single diode five-parameter model as the benchmark model.

[0022] Output current estimate of virtual photovoltaic array The calculation formula is as follows:

[0023] ,

[0024] : Number of series and parallel components;

[0025] Photocurrent, which is related to the measured irradiance G and temperature T: ;

[0026] I and V: Measured current and voltage of the photovoltaic array, respectively;

[0027] The reverse saturation current of a diode changes nonlinearly with temperature.

[0028] Equivalent series resistance, characterizing line loss and contact resistance;

[0029] Equivalent parallel resistance, characterizing leakage current, is typically used to detect hot spots or insulation faults;

[0030] n: Diode ideality factor, which depends on the module manufacturing process and is typically between 1.0 and 1.5;

[0031] Thermal voltage, defined as Where k is the Boltzmann constant and q is the electron charge. This refers to the absolute temperature of the battery.

[0032] The digital twin model for lithium battery energy storage adopts a second-order RC equivalent circuit model; the model includes a state integration stage of SOC.

[0033] The state-space equations are as follows:

[0034] ,

[0035] Output equation:

[0036] ,

[0037] s is the battery state of charge (SOC), with a value range of [0, 1].

[0038] These are the electrochemical polarization voltage and the concentration polarization voltage, respectively.

[0039] Battery load current,

[0040] Battery rated capacity,

[0041] Coulomb efficiency

[0042] Short-time constant RC network parameters characterize the charge transfer process.

[0043] Long-time constant RC network parameters characterize the lithium-ion diffusion process.

[0044] Internal resistance in ohms is a key indicator for judging battery aging, state of equilibrium (SOH), and poor contact.

[0045] Open-circuit voltage, which has a non-linear mapping relationship with SOC, is obtained through table lookup or polynomial fitting.

[0046] A DC / DC digital twin model of the power converter is established, and an average switching cycle model of the DC / DC Boost converter connecting the photovoltaic power generation unit and the DC bus is created.

[0047] Considering the ZIP integrated characteristics of the load, the load effect is manifested in the current drawn from the DC bus. ;

[0048] The state equation is:

[0049] ,

[0050] ,

[0051] D: Duty cycle,

[0052] Inductor current estimate

[0053] Photovoltaic array output voltage,

[0054] DC bus voltage

[0055] DC bus load current

[0056] Inductance and capacitance values

[0057] : Equivalent series resistance of inductance.

[0058] Based on the above system, a fault prediction method for a source-grid-load-storage coordinated system based on digital twins is implemented, including the following process:

[0059] Step S1: Data cleaning and preprocessing; The system receives raw data from the MQTT queue, performs validity checks, and removes null values ​​or garbled characters caused by packet loss; High-frequency noise is smoothed using a moving average filter, with the window size set to 5-10 sampling points;

[0060] Step S2: Digital twin state synchronization; input variables of the physical sensing layer, such as irradiance G and load current... The PWM duty cycle D is injected into the digital twin model; the hidden state of the twin model is corrected using an unscented Kalman filter (UKF).

[0061] Step S3: Generate residuals for the entire system and calculate the bias and residuals of key observations;

[0062] Define the original deviation as ,in These are measured values. For twin predictions, the residual magnitude is defined as... ;

[0063] Used for comparison with a threshold to trigger an alarm. The polarity is used for subsequent fault classification;

[0064] The table below lists the key residual definitions and their physical meanings in this embodiment:

[0065]

[0066] Step S4: Adaptive dynamic threshold calculation;

[0067] Dynamic gain is only enabled when changes in environmental sensors, such as a radiometer, or control commands cause power fluctuations.

[0068] The formula for calculating the dynamic threshold is as follows:

[0069] ,

[0070] Static baseline threshold; The standard deviation of the sensor, The usual value is 3;

[0071] The absolute value of the rate of change of the total power of the system, i.e.: ,

[0072] Dynamic sensitivity coefficient

[0073] : No. The time constant of each subsystem

[0074] Environmental factors, namely interlocking coefficients, are detected when there are drastic changes in irradiance G or load commands. If the external environment remains unchanged but the power suddenly changes, At this point, the threshold does not widen, ensuring that the fault is captured.

[0075] Safety margin;

[0076] Step S5: Fault mode classification and location;

[0077] When the residual amplitude And the duration exceeds When this occurs, the fault classifier is triggered;

[0078]

[0079] For open circuit of IGBT in energy storage converter; Symptoms: Increased DC side voltage ripple residual and increased AC side current waveform distortion rate (THD) residual;

[0080] Positioning: The ideal inverter in the digital twin outputs a pure sine wave. By comparing the FFT analysis results of the measured waveform, if the residual of a specific harmonic is significant, it is located as an open circuit of the IGBT in a specific bridge arm.

[0081] Step S6: Predictive maintenance recommendations are generated. The system not only outputs fault alarms, but also predicts the remaining lifespan based on parameter identification results.

[0082] Battery SOH prediction: If the identified internal resistance Given a linear trend of growth, time series forecasting models are used to extrapolate and predict when... The battery will reach its end-of-life (EOL) threshold, thus suggesting when to replace the battery clusters.

[0083] Compared with the prior art, the present invention has the following beneficial effects:

[0084] (i) Extremely high anti-interference capability. By introducing the dynamic derivative of power as a threshold adjustment factor, the problem of false alarms in digital twin tracking caused by intermittent fluctuations such as sudden changes in photovoltaic power output in new energy systems is effectively solved, and the robustness of the system is significantly improved.

[0085] (ii) Full life cycle monitoring. Utilizing physical mechanism models, the system can not only detect sudden failures, but also predict the aging trend and remaining life (RUL) of equipment by tracking model parameters such as the long-term drift of battery internal resistance, thereby achieving predictive maintenance.

[0086] (iii) No need for massive fault samples. This solution is based on deviation analysis of the mechanistic model, i.e., the "anomaly detection" approach. As long as the system deviates from the normal physical model trajectory, it is considered a potential fault. Therefore, it is not necessary to collect a large number of hard-to-obtain fault samples in advance for training, which greatly reduces the implementation threshold.

[0087] (iv) Accurate fault location. Through cross-verification of multiple physical quantities, for example: if the current residual is large but the voltage residual is small, it is determined to be a sensor fault; if the current and voltage are abnormal at the same time and the deviation conforms to Ohm's law, it is determined to be a component parameter drift, thus achieving precise decoupling of the fault source. Attached Figure Description

[0088] Figure 1 This is an overall architecture diagram of a source-grid-load-storage collaborative system fault prediction system based on digital twins, according to the present invention. It shows the physical entities, data transmission channels, digital twins of four model modules including source, grid, load, and storage, and the connection relationships of upper-layer application services.

[0089] Figure 2 This is a flowchart of the adaptive dynamic threshold determination logic of the present invention. It details the complete closed-loop logic from data input, residual calculation, calculation, environmental factor determination, dynamic threshold adjustment to fault output. Detailed Implementation

[0090] A fault prediction system for a source-grid-load-storage collaborative system based on digital twins includes: a physical sensing layer, a data interaction layer, and a digital twin service layer.

[0091] The physical sensing layer is the hardware entity of the source-grid-load-storage system, including the following subsystems and key sensor configurations:

[0092] The source side, i.e., the photovoltaic power generation unit, includes: a distributed photovoltaic array, a high-precision Hall current sensor deployed at the combiner box, and a voltage divider resistor network; the Hall current sensor has a sampling frequency greater than 1kHz and is deployed on the array plane to collect irradiance. Total radiation meter and used for collecting backplate temperature Surface mount sensor;

[0093] The energy storage side, i.e., the lithium battery energy storage unit, includes: a lithium iron phosphate battery pack equipped with a BMS; and it collects the voltage of individual battery cells. The second voltage sensor collects the total current of the battery cluster. The second current sensor collects the temperature at key points. Temperature sensor; BMS broadcasts data externally via CAN bus.

[0094] The grid side, i.e., power conversion and interface, includes: a bidirectional DC / AC inverter connecting the DC bus and the AC grid, and a device for measuring the three-phase voltage on the AC side. The third voltage sensor measures the three-phase current on the AC side. The third current sensor measures the DC bus voltage. The fourth voltage sensor.

[0095] The load side, or flexible load, includes: controllable lighting loads, electric vehicle charging stations; and real-time power data from smart meters. Power sensor, power factor acquisition The power factor acquisition device.

[0096] The data interaction layer includes: an industrial edge computing gateway based on ARM architecture, which collects data from each subsystem through bus polling; a communication protocol, which uses the MQTT protocol to upload the cleaned data to the digital twin server; and an NTP or PTP protocol to ensure that the data from the source, network, load, and storage sides have a unified timestamp, which is the basis for calculating instantaneous residuals.

[0097] The digital twin service layer is deployed on a high-performance server or cloud platform and runs the digital twin model; it includes: a numerical integral solver written in Python or C++, such as the Runge-Kutta 4th order method, for real-time solving of the differential equation system described below; and a fault analyzer that executes adaptive threshold calculation and fault logic judgment algorithms.

[0098] The digital twin model uses the state-space method to establish a unified system model; the photovoltaic power generation unit uses a single diode five-parameter model as the benchmark model.

[0099] Output current estimate of virtual photovoltaic array The calculation formula is as follows:

[0100] ,

[0101] : Number of series and parallel components;

[0102] Photocurrent, which is related to the measured irradiance G and temperature T: ;

[0103] I and V: Measured current and voltage of the photovoltaic array, respectively;

[0104] The reverse saturation current of a diode changes nonlinearly with temperature.

[0105] Equivalent series resistance, characterizing line loss and contact resistance;

[0106] Equivalent parallel resistance, characterizing leakage current, is typically used to detect hot spots or insulation faults;

[0107] n: Diode ideality factor, which depends on the module manufacturing process and is typically between 1.0 and 1.5;

[0108] Thermal voltage, defined as Where k is the Boltzmann constant and q is the electron charge. This refers to the absolute temperature of the battery.

[0109] Implementation details: Since the above formula is about The implicit function, i.e. the transcendental equation, is solved numerically by the digital twin engine using the Newton-Raphson method at each time step. It usually converges after 3-5 iterations, thus ensuring real-time performance.

[0110] Digital twin model of lithium battery energy storage:

[0111] Fault prediction for energy storage batteries primarily relies on the estimation of unobservable quantities of their internal states. This embodiment employs a second-order RC equivalent circuit model; to achieve closed-loop simulation, the model includes a state integration element for the state of charge (SOC).

[0112] The state-space equations are as follows:

[0113] ,

[0114] Output equation:

[0115] ,

[0116] s is the battery state of charge (SOC), with a value range of [0, 1].

[0117] These are the electrochemical polarization voltage and the concentration polarization voltage, respectively.

[0118] Battery load current,

[0119] Battery rated capacity,

[0120] Coulomb efficiency

[0121] Short-time constant RC network parameters characterize the charge transfer process.

[0122] Long-time constant RC network parameters characterize the lithium-ion diffusion process.

[0123] Internal resistance in ohms is a key indicator for judging battery aging, state of equilibrium (SOH), and poor contact.

[0124] Open-circuit voltage, which has a non-linear mapping relationship with SOC, is obtained through table lookup or polynomial fitting.

[0125] Parameter identification implementation: Online identification using recursive least squares method with forgetting factor. If identified If the number of cases rises sharply in a short period of time, the system will identify it as a "loose connection" or "broken tab" fault.

[0126] A DC / DC digital twin model of the power converter is established to create an average switching cycle model of the DC / DC Boost converter connecting the photovoltaic power generation unit and the DC bus; this model is also applicable to understanding and establishing the dynamic characteristics of other types of converters such as DC / AC.

[0127] Considering the ZIP integrated characteristics of the load, the load effect is manifested in the current drawn from the DC bus. ;

[0128] The state equation is:

[0129] ,

[0130] ,

[0131] D: Duty cycle,

[0132] Inductor current estimate

[0133] Photovoltaic array output voltage,

[0134] DC bus voltage

[0135] DC bus load current

[0136] Inductance and capacitance values

[0137] : Equivalent series resistance of inductance.

[0138] Based on the above system, a fault prediction method for a source-grid-load-storage coordinated system based on digital twins is implemented, including the following process:

[0139] Step S1: Data Cleaning and Preprocessing. The system receives raw data from the MQTT queue, performs validity checks, and removes null values ​​or garbled characters, such as NaN, caused by packet loss during communication. A moving average filter is used to smooth high-frequency noise, with the window size set to 5-10 sampling points.

[0140] Step S2: Digital twin state synchronization; input variables of the physical sensing layer, such as irradiance G and load current... The PWM duty cycle D is injected into the digital twin model; the hidden state of the twin model is corrected using an unscented Kalman filter (UKF).

[0141] Example: Assuming the initial SOC value of the battery model is incorrect, UKF will quickly pull the virtual SOC back to near the true value based on the difference between the measured voltage and the predicted voltage using the Kalman gain K, ensuring that subsequent fault prediction is based on the correct state starting point.

[0142] Step S3: Generate residuals for the entire system and calculate the bias and residuals of key observations.

[0143] Define the original deviation as ,in These are measured values. For twin predictions, the residual magnitude is defined as... ;

[0144] Used for comparison with a threshold to trigger an alarm. The polarity is used for subsequent fault classification;

[0145] The table below lists the key residual definitions and their physical meanings in this embodiment:

[0146]

[0147] Step S4: Adaptive dynamic threshold calculation.

[0148] In a power generation, grid, load, and energy storage system, power fluctuations are common. To distinguish between "normal dynamic deviations" and "fault deviations," this invention designs a dynamic threshold that changes with the system state. .

[0149] To prevent faults themselves, such as short circuits, from causing drastic power fluctuations that raise the threshold and mask the fault, this step adds a logic interlock: dynamic gain is only enabled when changes in environmental sensors, such as irradiance meters, or control commands cause power fluctuations.

[0150] The formula for calculating the dynamic threshold is as follows:

[0151] ,

[0152] Static baseline threshold; The standard deviation of the sensor, The usual value is 3;

[0153] The absolute value of the rate of change of the total power of the system, i.e.: ,

[0154] Dynamic sensitivity coefficient

[0155] : No. The time constant of each subsystem

[0156] Environmental factors, namely interlocking coefficients, are detected when... At this point, the threshold does not widen, ensuring that the fault is captured.

[0157] Safety margin;

[0158] Step S5: Fault mode classification and location.

[0159] When the residual amplitude And the duration exceeds When this occurs, the fault classifier is triggered;

[0160]

[0161] For open circuit of IGBT in energy storage converter; Symptoms: Increased DC side voltage ripple residual and increased AC side current waveform distortion rate (THD) residual;

[0162] Positioning: The ideal inverter in the digital twin outputs a pure sine wave. By comparing the FFT analysis results of the measured waveform, if the residual of a specific harmonic, such as the second harmonic, is significant, then the IGBT of that specific bridge arm is identified as open-circuited.

[0163] Step S6: Predictive maintenance recommendations are generated. The system not only outputs fault alarms but can also predict remaining lifespan based on parameter identification results.

[0164] Battery SOH prediction: If the identified internal resistance Given a linear trend of growth, time series forecasting models, such as ARIMA, are used for extrapolation to predict when... The battery will reach the EOL threshold, which is 200% of the initial value, thus suggesting when to replace the battery cluster.

[0165] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0166] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fault prediction system for a source-grid-load-storage coordinated system based on digital twins, characterized in that, include: Physical sensing layer, data interaction layer, digital twin service layer; The physical sensing layer is the hardware entity of the source-grid-load-storage system, including the following subsystems and key sensor configurations: The source side, i.e., the photovoltaic power generation unit, includes: a high-precision Hall current sensor and a voltage divider resistor network deployed at the combiner box for a distributed photovoltaic array; a total radiation meter deployed on the array plane; and a meter for collecting backsheet temperature. Surface mount sensor; The energy storage side, i.e., the lithium battery energy storage unit, includes: a lithium iron phosphate battery pack equipped with a BMS; and it collects the voltage of individual battery cells. The second voltage sensor collects the total current of the battery cluster. The second current sensor collects the temperature at key points. Temperature sensor; BMS broadcasts data externally via CAN bus; The grid side, i.e., power conversion and interface, includes: a bidirectional DC / AC inverter connecting the DC bus and the AC grid, and a device for measuring the three-phase voltage on the AC side. The third voltage sensor measures the three-phase current on the AC side. The third current sensor measures the DC bus voltage. The fourth voltage sensor; The load side, or flexible load, includes: controllable lighting loads, electric vehicle charging stations; and real-time power data from smart meters. Power sensor, power factor acquisition The power factor acquisition device; The data interaction layer includes: an industrial edge computing gateway based on ARM architecture, which collects data from each subsystem through bus polling; a communication protocol, which uses the MQTT protocol to upload the cleaned data to the digital twin server; and an NTP or PTP protocol to ensure that the data from the source, network, load, and storage sides have a unified timestamp. The digital twin service layer is deployed on a high-performance server or cloud platform and runs the digital twin model; it includes: a numerical integral solver written in Python or C++ for real-time solving of the differential equations described below; and a fault analyzer that executes adaptive threshold calculation and fault logic judgment algorithms.

2. The fault prediction system for a source-grid-load-storage coordinated system based on digital twins as described in claim 1, characterized in that, The digital twin model uses the state-space method to establish a unified system model; the photovoltaic power generation unit uses a single diode five-parameter model as the benchmark model. Output current estimate of virtual photovoltaic array The calculation formula is as follows: ， : Number of linked items; Photocurrent and measured irradiance and temperature Related: ; These represent the measured current and voltage of the photovoltaic array, respectively. The reverse saturation current of a diode changes nonlinearly with temperature. Equivalent series resistance, characterizing line loss and contact resistance; Equivalent parallel resistance, characterizing leakage current, is typically used to detect hot spots or insulation faults; The diode ideality factor depends on the module manufacturing process and is typically between 1.0 and 1.

5. Thermal voltage, defined as ,in Boltzmann's constant, The amount of electron charge. This refers to the absolute temperature of the battery. The digital twin model for lithium battery energy storage adopts a second-order RC equivalent circuit model; the model includes a state integration stage of SOC. The state-space equations are as follows: Output equation (terminal voltage prediction): ， : Battery state of charge (SOC), value range [0, 1]; : These are the electrochemical polarization voltage and the concentration polarization voltage (state variables), respectively. Battery load current (specified as positive for discharge and negative for charging, unit: Ampere A). Battery rated capacity, Coulomb efficiency Short-time constant RC network parameters characterize the charge transfer process. Long-time constant RC network parameters characterize the lithium-ion diffusion process. Internal resistance in ohms is a key indicator for judging battery aging (SOH) and poor contact faults. Open-circuit voltage, which has a non-linear mapping relationship with SOC, is obtained through table lookup or polynomial fitting. Parameter identification implementation: Online identification using recursive least squares method with forgetting factor. If identified If the number of cases rises sharply in a short period of time, the system will identify it as a "loose connection" or "broken tab" fault. A digital twin model of the power converter (DC / DC) is established, and an average switching cycle model of the DC / DC Boost converter connecting the photovoltaic power generation unit and the DC bus is created. Considering the ZIP integrated characteristics of the load, the load effect is manifested in the current drawn from the DC bus. ; The state equation is: ， ， Duty cycle (control signal obtained from physical controller). Inductor current estimate Photovoltaic array output voltage, DC bus voltage DC bus load current (as interference input, calculated from the back-end inverter and ZIP load model). Inductance and capacitance values : Equivalent series resistance of inductance.

3. A fault prediction method for a source-grid-load-storage coordinated system based on digital twins, implemented on the basis of the system described in claim 2, characterized in that, The process includes the following: Step S1: Data cleaning and preprocessing; The system receives raw data from the MQTT queue, performs validity checks, and removes null values ​​or garbled characters caused by packet loss; High-frequency noise is smoothed using a moving average filter, with the window size set to 5-10 sampling points; Step S2: Synchronize the digital twin state; input variable irradiance of the physical perception layer Load current PWM duty cycle Injected into the digital twin model; the hidden states of the twin model are corrected using unscented Kalman filtering (UKF); Step S3: Generate residuals for the entire system and calculate the bias and residuals of key observations; Define the original deviation as ,in These are measured values. For twin predictions, the residual magnitude is defined as... ; Used for comparison with a threshold to trigger an alarm. The polarity (positive / negative) is used for subsequent fault classification; The table below lists the key residual definitions and their physical meanings in this embodiment: ， Step S4: Adaptive dynamic threshold calculation; Dynamic gain is only activated when changes in environmental sensors or control commands cause power fluctuations. The formula for calculating the dynamic threshold is as follows: ， Static baseline threshold; This represents the standard deviation (noise level) of the sensor. The usual value is 3; The absolute value of the rate of change of the total power of the system ( ), Dynamic sensitivity coefficient : No. The time constant of each subsystem Environmental factors (interlock coefficients), when irradiance is monitored. Or when the load command changes drastically, If the external environment remains unchanged but the power suddenly increases (suspected electrical fault), At this point, the threshold does not widen, ensuring that the fault is captured. Safety margin; Step S5: Fault mode classification and location; When the residual amplitude And the duration exceeds When this occurs, the fault classifier is triggered; ， For open circuit of IGBT in energy storage converter; Symptoms: Increased DC-side voltage ripple residual and increased AC-side current waveform distortion rate (THD) residual; Positioning: The ideal inverter in the digital twin outputs a pure sine wave. By comparing the FFT analysis results of the measured waveform, if the residual of a specific harmonic is significant, it is located as an open circuit of the IGBT in a specific bridge arm. Step S6: Predictive maintenance recommendations are generated. The system not only outputs fault alarms, but also predicts the remaining lifespan based on parameter identification results. Battery SOH prediction: If the identified internal resistance Given a linear trend of growth, time series forecasting models are used to extrapolate and predict when... The battery will reach its end-of-life (EOL) threshold, thus suggesting when to replace the battery clusters.

Images