State estimation and control method for new energy box transformer

Abstract

The present application relates to the technical field of power system automation, and more particularly to a new energy box transformer state estimation and control method, which comprises the following steps: injecting a multi-frequency electrical excitation signal into a box-type substation and collecting voltage signals and current signals, establishing a digital twin model based on primary wiring topology and grounding network topology, performing joint state estimation and parameter estimation, obtaining a comprehensive feature vector representing the degradation state, and inputting a fault diagnosis and life assessment model to obtain a risk index; then constructing a risk weighted objective function with active power output, reactive power output and current as independent variables, and solving to obtain a control vector for adjusting power output, reactive power compensation and electrical protection device action setting value, thereby ensuring grid connection requirements while reducing operating risk and prolonging equipment life.

Description

Technical Field

[0001] This invention relates to the field of power system automation technology, and in particular to a method for estimating and controlling the state of new energy transformer substations. Background Technology

[0002] With a high proportion of new energy sources connected to the grid, prefabricated substations are widely used on the power generation side as the interface for data collection and grid connection. These prefabricated substations are subjected to high loads, frequent fluctuations, and outdoor environmental conditions for extended periods, which can lead to slow degradation of internal cables, joints, grounding devices, and transformer windings. Traditional operating methods mainly rely on periodic tests, manual inspections, and simple over-limit alarms, making it difficult to identify latent problems such as rising grounding resistance, increased joint contact resistance, and localized insulation deterioration in a timely manner.

[0003] Protection settings and power output are often configured based on static margins, which are out of sync with the actual degree of degradation. This leads to either long-term conservative operation resulting in underutilization of equipment capacity, or maintaining high output even as latent defects gradually accumulate, increasing the risk of faults and power outages. With the continuous expansion of renewable energy installations, how to achieve fine-grained sensing of internal electrical parameters, degradation mechanisms, and operational risks in prefabricated substations using existing measurement channels and control methods without significantly increasing hardware investment, and directly feed the results back to power regulation and protection configurations, has become a crucial issue for improving grid security and renewable energy absorption capacity. Summary of the Invention

[0004] To address the numerous problems existing in the prior art, this invention provides a method for state estimation and control of new energy prefabricated substations. This invention injects multi-frequency electrical excitation signals into the new energy prefabricated substation, establishes a digital twin model by combining the primary wiring and grounding network topology, obtains electrical parameters and voltage and current distributions using joint state estimation and parameter estimation, constructs a comprehensive feature vector representing the degradation state, drives the fault diagnosis and life assessment model to generate risk indicators, and finally optimizes the control vector under the risk-weighted objective of active power output, reactive power output, and current, thereby achieving unified coordination of power generation output, equipment health, and protection settings.

[0005] This specification provides one or more embodiments of a method for estimating and controlling the state of a new energy transformer substation, including the following steps:

[0006] Injecting multi-frequency electrical excitation signals into the new energy prefabricated substation and simultaneously acquiring voltage and current signals, the excitation response observation vector and operating condition vector are obtained.

[0007] A digital twin model is established based on the primary wiring topology and the grounding grid topology. Joint state estimation and parameter estimation are performed using the excitation response observation vector and the operating condition vector to obtain the parameter estimation vector and the state estimation vector.

[0008] A comprehensive feature vector representing the degradation state is constructed using parameter estimation vector, state estimation vector, and operating condition vector. The comprehensive feature vector is then input into the fault diagnosis and life assessment model to obtain risk indicators.

[0009] Based on risk indicators, a risk-weighted objective function is constructed with active power output, reactive power output, and current as independent variables. Under the conditions of satisfying power constraints, voltage constraints, current constraints, and grounding safety constraints, the control vector is obtained. The control vector is used to adjust the power output and the action setpoints of electrical protection devices in the new energy box-type substation, and to update the injection configuration of multi-frequency electrical excitation signals.

[0010] According to the method described in one or more embodiments of this specification, the multi-frequency electrical excitation signal is obtained by superimposing a set of discrete frequency sinusoidal electrical excitation signals. The voltage signal and current signal are sampled synchronously under a unified time reference. The amplitude and phase of the voltage signal and current signal at each discrete frequency are organized into an excitation response observation vector through frequency domain analysis. The operating condition vector includes the active power, reactive power and current RMS value obtained statistically between adjacent excitation cycles.

[0011] According to the method described in one or more embodiments of this specification, the digital twin model represents the primary equipment terminals and grounding points of the new energy box-type substation as graph nodes, and the conductors, transformer windings and grounding leads as graph edges. Resistance parameters, inductance parameters, capacitance parameters and grounding resistance parameters are set on each graph edge. The digital twin model uses the resistance parameters, inductance parameters, capacitance parameters and grounding resistance parameters in the parameter estimation vector as the electrical parameters of each graph edge, the voltage variables in the state estimation vector as the graph node voltages, and the current variables in the state estimation vector as the graph edge currents.

[0012] According to the method described in one or more embodiments of this specification, the joint state estimation and parameter estimation employ the extended Kalman filter algorithm and the recursive least squares algorithm. The extended Kalman filter algorithm uses the excitation response observation vector and the operating condition vector as the observation to update the state estimation vector, and the recursive least squares algorithm uses the state estimation vector as the regression variable and the excitation response observation vector as the observation to update the parameter estimation vector.

[0013] According to one or more embodiments of this specification, the method for constructing a comprehensive feature vector representing a degradation state using parameter estimation vector, state estimation vector, and operating condition vector includes: calculating a predicted excitation response observation vector using a digital twin model; constructing a residual signal using the difference between the excitation response observation vector and the predicted excitation response observation vector; and concatenating the residual signal with the change characteristics of the parameter estimation vector, state estimation vector, and operating condition vector within a preset time window to obtain a comprehensive feature vector.

[0014] According to one or more embodiments of this specification, the fault diagnosis and life assessment model includes a fault mode classification model and a life prediction model. The fault mode classification model takes a comprehensive feature vector as input and outputs the probability values ​​of grounding degradation fault mode, joint overheating fault mode and insulation aging fault mode. The life prediction model takes the comprehensive feature vector and the fault mode probability values ​​as input and outputs the remaining service life of each fault mode.

[0015] According to the method described in one or more embodiments of this specification, the risk index is a scalar risk measure obtained by weighting the failure mode probability value output by the failure mode classification model and the remaining useful life of the corresponding failure mode output by the lifetime prediction model according to preset weights.

[0016] According to the method described in one or more embodiments of this specification, the risk-weighted objective function takes the predicted trajectories of active power output, reactive power output, and current as independent variables within a finite prediction time range. It sums the deviation costs of active power output and reactive power output from the power planning curve with the weighted costs of current on the risk index, obtains the control vector corresponding to each prediction time through rolling optimization, and executes the control vector corresponding to the current time in each control cycle.

[0017] According to the method described in one or more embodiments of this specification, the control vector includes the active power reference value, reactive power reference value, and switching status of the capacitor bank and reactor connected to the grid-connected inverter of the new energy box-type substation. The control vector is sent to the grid-connected inverter and switching equipment through the field control device to realize power regulation and reactive power compensation.

[0018] According to the method described in one or more embodiments of this specification, the operating settings of the electrical protection device include overcurrent protection settings and ground fault protection settings. When adjusting the operating settings of the electrical protection device, the amplitude is limited according to the preset upper and lower limits, and the maximum allowable change between adjacent control cycles is restricted.

[0019] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows:

[0020] By injecting multi-frequency electrical excitation signals into the new energy prefabricated substation and establishing a digital twin model in conjunction with the primary wiring and grounding grid topology, online identification of grounding resistance, line parameters, and spatial distribution of voltage and current is achieved on the basis of existing acquisition channels, overcoming the limitations of traditional solutions that can only rely on offline tests or rough measurements.

[0021] By constructing a comprehensive feature vector using parameter estimation vector, state estimation vector, and operating condition vector, and introducing a fault mode classification model and a life prediction model, it is possible to distinguish different degradation mechanisms such as grounding degradation, joint overheating, and insulation aging, and to quantitatively assess the remaining service life. This solves the problem that existing technologies cannot directly convert monitoring quantities into executable maintenance decisions.

[0022] By constructing a risk-weighted objective function based on risk indicators, with active power output, reactive power output, and current prediction trajectory as independent variables, a coordinated trade-off between output plan tracking and equipment health risks is achieved under power constraints, voltage constraints, current constraints, and grounding safety constraints. This makes up for the shortcomings of traditional scheduling, which only adjusts according to the planned curve or a single safety margin.

[0023] By simultaneously applying control vectors to the active and reactive power reference values ​​of the grid-connected inverter, the switching status of capacitor banks and reactors, and the operating settings of electrical protection devices, and jointly updating the multi-frequency electrical excitation signal injection configuration, integrated closed-loop control of state estimation, risk assessment, power regulation, and protection setting is achieved. Compared with the existing configuration method where each subsystem is independent, this improves the overall operational consistency and automation level. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the execution flow of the method of the present invention. Detailed Implementation

[0025] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation.

[0026] The "state" of a new energy prefabricated substation is not simply a matter of whether voltage or current exceeds limits, but a comprehensive operational profile composed of primary equipment parameters, grounding grid health, conductor and joint heating levels, insulation aging, and the current interaction between active and reactive power output and the power grid. This state includes the electrical quantity distribution at a given moment, the trend of slow parameter drift, and sensitivity to potential faults under different operating conditions. Relying solely on traditional measurement points and fixed thresholds makes it difficult to accurately answer key questions such as "Is the equipment currently safe?", "How much output can it still operate at?", and "When is maintenance required?". Therefore, this invention focuses on the observability, quantification, and controllability of the state of new energy prefabricated substations. By introducing multi-frequency electrical excitation, digital twin models, degradation feature construction, and risk-driven control, the state estimation results are directly transformed into decision-making basis for output adjustment and protection configuration, achieving dynamic characterization and closed-loop control of the state of new energy prefabricated substations.

[0027] like Figure 1 As shown, a method for estimating and controlling the state of a new energy transformer substation includes the following steps:

[0028] Injecting multi-frequency electrical excitation signals into the new energy prefabricated substation and simultaneously acquiring voltage and current signals, the excitation response observation vector and operating condition vector are obtained.

[0029] The multi-frequency electrical excitation signal is obtained by superimposing a set of discrete frequency sinusoidal electrical excitation signals. The voltage and current signals are sampled synchronously under a unified time base. The amplitude and phase of the voltage and current signals at each discrete frequency are organized into an excitation response observation vector through frequency domain analysis. The operating condition vector includes the active power, reactive power and current RMS values ​​obtained statistically between adjacent excitation cycles.

[0030] In this embodiment, a multi-frequency electrical excitation signal is introduced into the new energy prefabricated substation under normal grid connection conditions to obtain response data beneficial for state estimation. Specifically, a dedicated excitation device is installed on the low-voltage bus side or zero-sequence circuit side of the prefabricated substation. The excitation device is connected to the primary equipment through an isolation transformer or injection coupling device, and injects a multi-frequency electrical excitation signal into the target circuit within a pre-set excitation time period. The multi-frequency electrical excitation signal is formed by superimposing several discrete frequency sinusoidal signals. Each discrete frequency is selected outside the power frequency and its harmonic frequencies, and the amplitude is adjusted according to the rated voltage and rated current of the prefabricated substation and the power quality limits, so that the additional signal is relatively small compared to the power frequency, does not cause malfunction of the protection device, and does not cause significant disturbance to the upstream power grid. The excitation time period is usually arranged during a period of relatively stable load. The frequency and amplitude of the multi-frequency electrical excitation signal remain unchanged within each excitation time period, which facilitates horizontal comparison between different excitation time periods.

[0031] To acquire the voltage and current signals corresponding to the multi-frequency electrical excitation signals, multi-channel synchronous measurement devices are configured at key nodes in the prefabricated substation. Voltage transformers are used to collect bus voltage, and current transformers are used to collect current in each circuit. Their secondary side signals are connected to the data acquisition unit through an anti-aliasing filter. The data acquisition unit uses a unified clock source to synchronously sample all voltage and current channels at the same sampling frequency. The sampling start time is aligned with the injection time of the multi-frequency electrical excitation signal, and voltage and current samples covering several power frequency cycles are continuously acquired within each excitation time period. Synchronous sampling ensures that the phase relationship between different measurement points is consistent on the time axis, providing a basis for subsequent frequency domain analysis to recover the relative phase of each measurement point at each discrete frequency.

[0032] After sampling for an excitation period, the data processing unit performs frequency domain analysis on the voltage and current samples within that period. The processing flow includes truncating samples within the stable injection interval of the excitation period, applying a window function to each measurement channel to suppress spectral leakage, and then performing a Discrete Fourier Transform on each channel to calculate the voltage amplitude, phase, and current amplitude and phase at each discrete frequency of the multi-frequency electrical excitation signal. For the same discrete frequency, the voltage amplitude, voltage phase, current amplitude, and current phase from different measurement points are arranged in a predetermined order to form a set of observation data. Subsequently, the observation data corresponding to each discrete frequency are sequentially concatenated in one dimension to form the excitation response observation vector corresponding to that excitation period. The excitation response observation vector centrally characterizes the amplitude-frequency and phase-frequency characteristics of the prefabricated substation under multi-frequency excitation and is one of the main inputs for subsequent parameter estimation and state estimation.

[0033] While constructing the excitation response observation vector, this embodiment also constructs an operating condition vector based on historical measurements from the monitoring system to characterize the load status of the prefabricated substation between adjacent excitation time periods. Specifically, between two adjacent excitation time periods, the active power measurement value, reactive power measurement value, and current measurement value of each circuit of the prefabricated substation are read from the monitoring system at fixed time intervals. Within a time window corresponding to a certain excitation time period, the active power measurement value and reactive power measurement value are averaged, and the effective current value of each circuit current measurement value is calculated. These are then combined into an operating condition vector according to a predetermined order. The operating condition vector corresponds one-to-one with the excitation response observation vector. The former reflects the long-term load level and power factor within the excitation time period, while the latter reflects the local response of the prefabricated substation to multi-frequency electrical excitation signals under the load background. Both are used together to drive the state estimation and parameter estimation algorithms of the subsequent digital twin model, achieving a detailed characterization of the operating status of the new energy prefabricated substation. Through the above data acquisition and vector construction methods, this embodiment can be implemented relying only on existing measurement devices and conventional signal processing algorithms, demonstrating good engineering feasibility.

[0034] A digital twin model is established based on the primary wiring topology and the grounding grid topology. Joint state estimation and parameter estimation are performed using the excitation response observation vector and the operating condition vector to obtain the parameter estimation vector and the state estimation vector.

[0035] In this embodiment, a digital twin model is first constructed based on the primary wiring topology and grounding grid topology of the new energy prefabricated substation. The primary wiring topology can be derived from design drawings or operation monitoring systems, with busbars, high and low voltage sides of transformers, and feeder terminals as electrical nodes, and cables, busbar segments, transformer windings, and switchgear as connecting branches. The grounding grid topology is described using grounding down conductors, grounding electrodes, fences, and other grounding nodes and branches. In this embodiment, the digital twin model is an equivalent circuit model for the new energy prefabricated substation, capable of calculating the voltage of each node, the current of each branch, and the response to multi-frequency electrical excitation signals under given operating conditions and electrical parameters.

[0036] In a digital twin model, the parameter estimation vector represents the electrical parameters that need to be identified online, such as cable resistance, transformer winding resistance, grounding resistance, and some reactance and capacitance parameters. Initial values ​​can be provided based on nameplate parameters, design parameters, or experimental parameters. The state estimation vector represents state variables such as node voltages and branch currents under current operating conditions. Initial states can be provided by monitoring system measurements or power flow calculations. Through this partitioning, the digital twin model can update slowly changing electrical parameters while maintaining structural integrity and tracking instantaneous operating states.

[0037] To clarify the relationship between the digital twin model and the measured quantity in the algorithm implementation, this embodiment uses the following predictive relation to describe the mapping relationship between the model output and the input:

[0038]

[0039] in, The predicted stimulus response observation vector output by the digital twin model. For the parameter estimation vector, This is the state estimation vector. This represents the operating condition vector. (Function) For circuit calculation functions established based on primary wiring topology and grounding grid topology, given parameter estimation vector, state estimation vector, and operating condition vector, the function... The node voltages and branch currents at each discrete frequency are obtained by solving the network equations, and then converted into amplitude and phase forms consistent with the excitation response observation vector.

[0040] In the joint state estimation and parameter estimation process, the control backend receives the corresponding excitation response observation vector and operating condition vector in each excitation cycle, and performs a round of prediction and correction by combining the parameter estimation vector and state estimation vector obtained in the previous cycle. In the prediction phase, the state estimation vector is updated based on the operating condition vector to reflect the impact of load changes on voltage and current, while keeping the parameter estimation vector unchanged. In the correction phase, the predicted digital twin model output is compared with the actual excitation response observation vector to form a response error signal. The state estimation vector and parameter estimation vector are then jointly corrected based on this response error signal. This embodiment can employ conventional online estimation algorithms such as extended Kalman filtering combined with recursive least squares to implement the above prediction and correction process. The algorithm implementation can be completed using publicly available state estimation techniques.

[0041] In implementation, the joint state estimation algorithm outputs an updated state estimation vector in each iteration, representing the estimated values ​​of the node voltage and branch current under the current excitation cycle. Simultaneously, it outputs an updated parameter estimation vector, representing the current estimated values ​​of slowly changing electrical parameters such as grounding resistance and joint resistance. As the excitation cycle progresses, the parameter estimation vector gradually converges to values ​​closer to the actual operating conditions, thereby improving the accuracy of the digital twin model. The state estimation vector, on the other hand, can provide relatively complete and consistent station-wide state information even when monitoring data is incomplete, local sensor failures occur, or measurement noise is high.

[0042] By establishing a digital twin model based on the primary wiring topology and grounding grid topology, and using joint state estimation and parameter estimation based on excitation response observation vectors and operating condition vectors, this embodiment achieves online identification of the internal state and key electrical parameters of new energy prefabricated substations, building upon existing measurement methods. This ensures consistency between the model and the actual equipment structure while providing high-quality input data for subsequent degradation diagnosis and risk assessment. The computations required for this process can be performed on conventional industrial computers or edge computing units, facilitating deployment and maintenance at engineering sites.

[0043] The digital twin model represents the primary equipment terminals and grounding points of the new energy prefabricated substation as graph nodes, and the conductors, transformer windings and grounding leads as graph edges. Resistance parameters, inductance parameters, capacitance parameters and grounding resistance parameters are set on each graph edge. The digital twin model uses the resistance parameters, inductance parameters, capacitance parameters and grounding resistance parameters in the parameter estimation vector as the electrical parameters of each graph edge, the voltage variables in the state estimation vector as the graph node voltages, and the current variables in the state estimation vector as the graph edge currents.

[0044] The digital twin model models the primary equipment structure and grounding grid structure of a new energy prefabricated substation. First, busbars, transformer high and low voltage terminals, circuit breaker terminals, feeder outgoing terminals, and grounding points are extracted from design drawings or primary wiring diagrams. Each electrical connection terminal and grounding point is numbered as a node, forming a node list. Then, the conductors connecting these terminals and grounding points, transformer windings, busbar segments, and grounding down conductors are treated as branches. Each branch records its start node, end node, and equipment type, forming a branch list. Using the node and branch lists, the topology of the prefabricated substation's primary equipment and grounding grid can be fully described graphically in the control system.

[0045] For each branch, the digital twin model sets resistance, inductance, capacitance, and grounding resistance parameters at the parameter level. Initial values ​​can be provided by design calculations or factory test data; for example, conductor resistance is calculated from length and resistance per unit length, transformer winding resistance is given by DC resistance test results, and grounding resistance of the grounding lead is given by grounding resistance test results. The parameter estimation vector stores the resistance, inductance, capacitance, and grounding resistance parameters of each branch sequentially according to their branch numbers, thus establishing a one-to-one correspondence between the vector positions and the physical branches, facilitating subsequent online updates.

[0046] The state estimation vector records the estimated values ​​of the voltage at each node and the current in each branch under the current operating conditions. In the digital twin model, each node voltage component corresponds to a node number, and each branch current component corresponds to a branch number. The state estimation vector output by the state estimation algorithm is directly written into the digital twin model, so that node objects hold the corresponding voltage estimates, and branch objects hold the corresponding current estimates. In this way, the digital twin model has a complete "total station voltage" and "total station current" estimation distribution at any given time, which can provide input for subsequent degradation diagnosis and risk assessment.

[0047] To transform the parameter estimation vector and state estimation vector into an electrical model usable for response calculation, the digital twin model employs the conventional nodal admittance method to construct the network equations. For any excitation frequency, the corresponding complex impedance is calculated based on the resistance, inductance, capacitance, and grounding resistance parameters of each branch, then converted into branch admittance, and assembled into a nodal admittance matrix according to the node connection relationships. This process can be briefly represented by the following relationship:

[0048]

[0049] in, The branch current vector, For node voltage vectors, The nodal admittance matrix is ​​constructed from the parameter estimation vectors. This is the parameter estimation vector. Given the power injection conditions determined by the operating condition vector, the node voltage and branch current distributions at the corresponding frequency can be obtained by solving the above relationship. The calculated node voltage and branch current correspond structurally one-to-one with the voltage and current variables in the state estimation vector, and can be used to generate predictive excitation responses or to verify the consistency of the state estimation results.

[0050] At the implementation level, the control system can store the node and branch information of the graph structure in a database or in-memory data structure. Each component of the parameter estimation vector points to the resistance, inductance, capacitance, and grounding resistance parameters of a specific branch through an index, and each component of the state estimation vector points to the voltage of a specific node or the current of a branch through an index. When the joint state estimation algorithm outputs new parameter and state estimation vectors, only the position of the corresponding index needs to be updated, and the digital twin model can use the latest parameters and states in the next solution. Through the above-described graph structure representation based on nodes and branches, this embodiment achieves a clear mapping between parameter estimation vectors, state estimation vectors, and physical devices while maintaining a high degree of consistency between the model structure and the actual primary equipment. This allows those skilled in the art to directly build and solve the digital twin model based on existing power system analysis software or a self-developed power flow calculation module.

[0051] The joint state estimation and parameter estimation adopt the extended Kalman filter algorithm and the recursive least squares algorithm. The extended Kalman filter algorithm uses the excitation response observation vector and the operating condition vector as the observation to update the state estimation vector, while the recursive least squares algorithm uses the state estimation vector as the regression variable and the excitation response observation vector as the observation to update the parameter estimation vector.

[0052] The purpose of joint state estimation and parameter estimation is to simultaneously obtain online estimates of the current operating state and key electrical parameters of the new energy prefabricated substation while the excitation response observation vector and operating condition vector are continuously updated. The state estimation vector is used to characterize quantities that change rapidly with the load, such as the voltage of each node and the current of each branch, while the parameter estimation vector is used to characterize electrical parameters that drift slowly over time, such as cable resistance, transformer winding resistance, and grounding resistance. Since there is a significant difference between the two in terms of time scale, this embodiment uses the extended Kalman filter algorithm for state estimation and the recursive least squares algorithm for parameter estimation, and connects the two into a closed-loop process through the excitation response observation vector and the operating condition vector.

[0053] To map the observation information to the digital twin model, the joint estimation process treats the excitation response observation vector obtained under multi-frequency excitation and the operating condition vector as both the output and input of the measurement equation. The measurement equation can be abstractly written as:

[0054]

[0055] in, For the excitation response observation vector, This is the state estimation vector. For the parameter estimation vector, For the operating condition vector, the function This is the nonlinear response relationship given by the digital twin model. This relationship provides a model expectation of "what kind of excitation response should be produced under the current state and parameters under given operating conditions," providing a unified physical constraint for the extended Kalman filter algorithm and the recursive least squares algorithm.

[0056] The Extended Kalman Filter (EKF) algorithm updates the state estimation vector using a "prediction-correction" approach. Specifically, at the beginning of each excitation cycle, the state estimation vector and parameter estimation vector from the previous excitation cycle are combined with the current operating condition vector to perform a forward calculation in the digital twin model, obtaining a predicted value for the current state. The prediction error covariance is then set based on experience or offline analysis. After the excitation cycle ends, the corresponding excitation response observation vector is acquired. The predicted model response is compared with the excitation response observation vector, the observation residual is calculated, and the observation error covariance is constructed based on the measurement noise level. The EKF algorithm performs first-order linearization of the measurement equation near the current operating point, obtains the gain coefficient, and uses the observation residual to correct the state estimation vector, thereby obtaining the state estimation vector for the current excitation cycle. This ensures that the digital twin model output is as close as possible to the actual excitation response observation vector.

[0057] The recursive least squares algorithm is used to slowly adjust the parameter estimation vector. Within each excitation cycle, after the extended Kalman filter algorithm obtains a new state estimation vector, the digital twin model can calculate a set of theoretical excitation responses under this state estimation vector and the operating condition vector. Subtracting the observed excitation response vector from the theoretical excitation response yields the parameter-related residual signal. Utilizing the sensitive relationship between the residual signal and the theoretical excitation response, a regression vector is constructed. The component in the state estimation vector related to parameter changes is used as the regression variable, and the residual signal is used as the input to the recursive least squares algorithm. The recursive least squares algorithm incrementally updates the parameter estimation vector once per excitation cycle and controls the weight of old and new information through a forgetting factor, ensuring that electrical parameters can be identified during long-term drift without being overly sensitive to short-term noise.

[0058] In its implementation, the control system iterates according to the excitation cycle: First, the state estimation vector and parameter estimation vector from the previous cycle are used as initial values. These are then combined with the current operating condition vector to complete the prediction and correction steps of the extended Kalman filter algorithm, resulting in a new state estimation vector. Subsequently, based on the new state estimation vector and the digital twin model, the theoretical excitation response is calculated, forming a residual signal that drives the recursive least squares algorithm to update the parameter estimation vector. The updated parameter estimation vector participates in the prediction step of the extended Kalman filter algorithm in the next excitation cycle, thus forming a linkage feedback between state estimation and parameter estimation. Through this clearly defined and rhythmically matched joint estimation structure, this embodiment can maintain the sensitivity of the state estimation vector to the real-time operating state under conditions of multi-frequency excitation and complex changes in operating conditions, while ensuring the convergence and stability of the parameter estimation vector in long-term operation, providing reliable basic data for subsequent degradation analysis and control decisions.

[0059] A comprehensive feature vector representing the degradation state is constructed using parameter estimation vector, state estimation vector, and operating condition vector. The comprehensive feature vector is then input into the fault diagnosis and life assessment model to obtain risk indicators.

[0060] In this method, the parameter estimation vector, state estimation vector, and operating condition vector are not directly used in the downstream model. Instead, they are first organized into a comprehensive feature vector that can centrally reflect the equipment degradation process. Degradation states include a slow increase in grounding resistance, an increase in joint contact resistance, local overheating of windings, and an increase in insulation loss. These changes are often not obvious in a single measurement, but they have stable characteristics in the relationship between parameter change trends, operating load levels, and voltage and current distributions. Therefore, it is necessary to reorganize the three types of vectors in both time and physical meaning dimensions so that the comprehensive feature vector can simultaneously reflect the "current state," "historical evolution," and "current operating condition."

[0061] In practical implementation, the control system uses several consecutive excitation cycles as a time window. It extracts the current value, change amount, and rate of change of each key electrical parameter within that window from the parameter estimation vector. For example, it extracts the current estimated value of the grounding resistance, the difference between two adjacent excitation cycles, and the weighted average of the differences, to characterize long-term degradation trends. It also extracts bus voltage deviation, current imbalance in each loop, and phase-to-phase current ratio from the state estimation vector to characterize the stress and heat levels under the current operating conditions. Finally, it extracts the average active power, average reactive power, effective current value, and power factor from the operating condition vector to characterize load intensity. After normalization and encoding, these quantities are concatenated in a fixed order to form a comprehensive feature vector, represented by the following symbols:

[0062]

[0063] in, For the comprehensive feature vector, For the parameter estimation vector, This is the state estimation vector. For the operating condition vector, the function Features are used to construct functions, including time window selection, difference operations, normalization, and concatenation.

[0064] The fault diagnosis and lifetime assessment models use a comprehensive feature vector as a unified input interface and can be divided into two sub-models in engineering implementation. The fault diagnosis model is a multi-classification model used to identify which typical fault mode is most likely to occur, such as grounding degradation fault, joint overheating fault, or insulation aging fault, outputting the probability of occurrence for each fault mode. The lifetime assessment model is a regression model used to estimate the remaining lifetime of the corresponding fault mode given the fault mode probability and the comprehensive feature vector. Both models can employ mature algorithms such as random forests, gradient boosting trees, or feedforward neural networks, and can be trained offline using historical operating data, experimental data, and simulation data, enabling the models to learn the correspondence between the comprehensive feature vector and fault category and lifetime at different degradation stages.

[0065] During the online operation phase, after each joint state estimation and parameter estimation, a new comprehensive feature vector is constructed and fed into the fault diagnosis and lifetime assessment model to obtain the current fault mode probability and remaining lifetime estimate. To facilitate subsequent control decisions, this method summarizes these outputs into a scalar risk index to quantify the overall operational risk level of the current new energy prefabricated substation. The risk index can combine the "probability of a certain fault mode occurring" and the "length of the remaining lifetime of that fault mode" according to preset weights. The shorter the remaining lifetime and the more severe the fault mode, the greater the corresponding weight, and the higher the risk index. Through this diagnosis and assessment link centered on the comprehensive feature vector, this method closely integrates the digital twin estimation results with the degradation mechanism and operating conditions, enabling the risk index to truly reflect the health status and future evolution trend of the equipment under current operating conditions, providing reliable input for subsequent risk-weighted control.

[0066] The comprehensive feature vector representing the degradation state is constructed using parameter estimation vector, state estimation vector, and operating condition vector. This includes: calculating the predicted excitation response observation vector using a digital twin model; constructing a residual signal using the difference between the excitation response observation vector and the predicted excitation response observation vector; and concatenating the residual signal with the change characteristics of the parameter estimation vector, state estimation vector, and operating condition vector within a preset time window to obtain the comprehensive feature vector.

[0067] To ensure that the characterization of the degradation state relies not only on the numerical values ​​of the parameter estimation vector, state estimation vector, and operating condition vector, but also reflects the deviation between the model and the actual response, after each excitation cycle, the digital twin model calculates the corresponding predicted excitation response observation vector under the current parameter estimation vector, state estimation vector, and operating condition vector, and compares it with the actual excitation response observation vector to obtain the residual signal. The residual signal physically reflects "the part of the response that the model cannot explain under the current estimated parameters and states," and is typically associated with early degradation phenomena such as grounding degradation, poor joint contact, and localized insulation dampness.

[0068] The residual signal can be calculated as follows:

[0069]

[0070] in, The residual signal vector, For the excitation response observation vector, This is the predicted excitation response observation vector calculated from the digital twin model. The residual signal vector has a corresponding component at each frequency and each measurement channel, directly reflecting the magnitude of the model bias at that frequency and location.

[0071] To avoid the influence of instantaneous noise or individual abnormal sampling on the single residual signal, the system uses a preset time window composed of several consecutive excitation periods to perform statistical processing on the residual signal within the window. Specifically, the average value, maximum value, and range of variation of each dimension of the residual signal can be calculated to describe the stable offset and fluctuation of the residual in that dimension over time; the energy of the residuals at each frequency can also be summed to reflect the overall amplitude-frequency deviation level. The above statistical results are arranged in a fixed order to form a residual feature sub-vector.

[0072] Within the same time window, the parameter estimation vector is used to extract the difference between the current value and the initial value, the change and trend between adjacent excitation cycles, to characterize whether key parameters such as grounding resistance and winding resistance show a continuous increase or abrupt change within the window. The state estimation vector is used to extract bus voltage deviation, loop current imbalance, and peak current, to reflect whether the voltage and current distribution shows abnormal concentration or deviation under the current operating condition. The operating condition vector is used to extract the average active power, average reactive power, effective current value, and power factor, to characterize the thermal and electrical stress levels experienced by the prefabricated substation within this time window. After dimensional normalization and range clipping, these features are respectively used to form parameter feature sub-vectors, state feature sub-vectors, and operating condition feature sub-vectors.

[0073] The construction process of the comprehensive feature vector is as follows: First, within a time window, the residual feature vector, parameter feature vector, state feature vector, and operating condition feature vector are updated sequentially. Then, the four types of sub-vectors are concatenated in a one-dimensional manner according to a preset order to obtain the comprehensive feature vector used to characterize the degradation state. This comprehensive feature vector, within a unified data structure, simultaneously includes "the deviation between the model and the actual measurement," "the long-term changes in electrical parameters," "the spatial distribution characteristics of voltage and current," and "the current load condition," enabling a relatively complete depiction of the current health status and evolution trend of the new energy prefabricated substation. After receiving the comprehensive feature vector, the fault diagnosis and life assessment model can distinguish different degradation mechanisms such as grounding degradation, joint overheating, and insulation aging based on patterns learned from historical samples, and output risk assessment results that match the comprehensive feature vector.

[0074] The fault diagnosis and life assessment model includes a fault mode classification model and a life prediction model. The fault mode classification model takes the comprehensive feature vector as input and outputs the probability values ​​of grounding degradation fault mode, joint overheating fault mode and insulation aging fault mode. The life prediction model takes the comprehensive feature vector and fault mode probability values ​​as input and outputs the remaining service life of each fault mode.

[0075] In this embodiment, fault diagnosis and life assessment are implemented using two complementary models: a fault mode classification model and a life prediction model. Both models use the comprehensive feature vector constructed in the previous step as their main input. This comprehensive feature vector integrates parameter estimation vectors, state estimation vectors, operating condition vectors, and residual signal changes within a preset time window, enabling a relatively complete depiction of the health status evolution process of the new energy prefabricated substation within a certain time range.

[0076] Before building the model, a labeled training sample library needs to be constructed. Samples can be obtained from the following data sources: First, prefabricated substations that have experienced faults or serious defects such as grounding degradation, joint overheating, and insulation aging in actual operation, with the comprehensive feature vectors of the period before the fault occurrence mapped to the fault type; second, the test results of equipment scheduled for maintenance or decommissioning, such as grounding resistance test values, joint infrared temperature measurement results, and insulation test results, which are used as labels for the degree of degradation; third, simulation data based on digital twin models and experimental platforms, generating corresponding comprehensive feature vectors and "virtual fault modes" under different parameter degradation combinations. All types of samples are uniformly divided according to time windows, forming a training dataset of "comprehensive feature vector + fault mode label + remaining service life label".

[0077] The fault mode classification model takes a comprehensive feature vector as input and outputs the probability of occurrence of each predefined fault mode. Predefined fault modes include at least grounding degradation fault modes, joint overheating fault modes, and insulation aging fault modes; normal operation modes can also be added as needed. The model structure can employ mature classification algorithms such as random forests, gradient boosting trees, or feedforward neural networks. During the training phase, model parameters are adjusted by minimizing classification error, ensuring that samples with similar comprehensive feature vectors are classified into the same fault mode. During online operation, for each time window, the corresponding comprehensive feature vector is input into the fault mode classification model to obtain the probability value of each fault mode. Maintenance personnel can determine the most likely degradation type of the current equipment based on the probability values, or set a probability threshold to trigger an alert when the probability of a certain fault mode consistently exceeds the threshold.

[0078] The life prediction model further estimates the remaining service life based on the failure mode classification model. The input to the life prediction model consists of two parts: a comprehensive feature vector and the probability values ​​of each failure mode output by the failure mode classification model. The model output is the remaining service life corresponding to each failure mode. When training the life prediction model, the "operating time from the current time window to the actual occurrence of failure or reaching the ultimate degradation state" in historical samples can be used as life labels. A multi-output regression model is used to fit the correspondence between the comprehensive feature vector, failure mode probabilities, and life labels. To improve engineering applicability, the life prediction results can be given in the form of operating hours, operating days, or equivalent load cycles, facilitating direct integration with maintenance cycles and spare parts replacement plans.

[0079] During the online application phase, after each round of joint state estimation and parameter estimation, the system first constructs a comprehensive feature vector for the current time window. This vector is then input into the fault mode classification model to obtain probability values ​​for grounding degradation fault modes, joint overheating fault modes, and insulation aging fault modes. The comprehensive feature vector and the aforementioned probability values ​​are then input into the lifetime prediction model to obtain the remaining lifetime estimate for each fault mode. To reduce the impact of instantaneous noise, a moving average or median filter can be applied to the probability values ​​and lifetime estimates of multiple adjacent time windows, making the output results smoother over time and more consistent with the actual degradation process of the equipment. Through the above fault diagnosis and lifetime assessment models, this embodiment can not only identify the main degradation types that may currently exist in new energy prefabricated substations, but also provide the remaining lifetime for different degradation types, providing a clear and actionable basis for subsequent risk quantification and control decisions.

[0080] The risk index is a scalar risk measure obtained by weighting the failure mode probability value output by the failure mode classification model and the remaining useful life of the corresponding failure mode output by the lifetime prediction model according to preset weights.

[0081] In this invention, a scalar measure called a risk index is introduced to transform the outputs of the fault mode classification model and the lifetime prediction model into quantifiable quantities that can directly drive control and optimization. For each time window, the fault mode classification model outputs the probabilities of grounding degradation fault modes, joint overheating fault modes, and insulation aging fault modes, while the lifetime prediction model outputs the corresponding remaining lifetimes for grounding degradation fault modes, joint overheating fault modes, and insulation aging fault modes. The control system first standardizes the remaining lifetimes, assigning very short lifetimes to high-risk levels and longer lifetimes to low-risk levels. Then, combining the probability values ​​of each fault mode with the pre-set severity weights in the operation and maintenance strategy, a single scalar risk index is calculated to comprehensively characterize the operational risk of the new energy prefabricated substation within the current time window.

[0082] In terms of implementation, the following weighted calculation method can be adopted:

[0083]

[0084] in, As a risk indicator, For the first The weights of each failure mode, For the first The probability value of each failure mode. For the first Remaining useful life for each failure mode To map the remaining service life as a monotonically decreasing function of a dimensionless risk factor, weights are pre-set according to the impact of different failure modes on personal safety, power grid stability, and equipment damage costs. Higher weights are assigned to grounding degradation and insulation aging failure modes, while relatively lower weights are assigned to joint overheating failure modes, making the risk index numerically more sensitive to high-risk failure modes.

[0085] To improve the stability of risk indicators over time, the control system performs a moving average smoothing process on the risk indicators calculated over multiple consecutive time windows, suppressing drastic fluctuations caused by single-estimate bias and measurement noise. When the smoothed risk indicator consistently exceeds a preset threshold, a tiered early warning is triggered. Different thresholds correspond to different maintenance response levels: medium-level warnings correspond to scheduled maintenance, while high-level warnings correspond to reducing renewable energy output or scheduling power outages for inspection. Through this construction method, the risk indicator unifies the probability of failure, degradation rate, and maintenance preferences into a single scalar dimension. This allows subsequent power scheduling and protection setting adjustments based on a risk-weighted objective function to be balanced around the same evaluation standard, maintaining renewable energy output levels while ensuring safety.

[0086] Based on risk indicators, a risk-weighted objective function is constructed with active power output, reactive power output, and current as independent variables. Under the conditions of satisfying power constraints, voltage constraints, current constraints, and grounding safety constraints, the control vector is obtained. The control vector is used to adjust the power output and the action setpoints of electrical protection devices in the new energy box-type substation, and to update the injection configuration of multi-frequency electrical excitation signals.

[0087] In this embodiment, the risk index, as the core constraint for the regulation of the new energy prefabricated substation, is directly embedded in the power optimization process. Within each control cycle, the control system constructs a risk-weighted objective function based on the predicted trajectories of current active power output, reactive power output, and current, unifying economic and safety objectives within the same optimization framework. Therefore, at the beginning of the control cycle, the scheduling module first reads the risk index, power plan curve, and current operating conditions of the prefabricated substation from the previous cycle. Combining this with a digital twin model, it performs rolling predictions of active power output, reactive power output, and current at each moment in the next prediction period, forming candidate trajectories for adjustable variables.

[0088] The risk-weighted objective function is expressed in the following form:

[0089]

[0090] in, The risk-weighted objective function is... This is the vector of active power output. This is the reactive power vector. It is a current vector. As a risk indicator, Let be the cost function for the deviation of active and reactive power output relative to the power plan curve. This is a risk cost function that shows the increasing cost as the current approaches the thermal stability limit at a given risk level. and These are weighting coefficients used to balance economic efficiency and safety. In practice, the deviation cost function can be expressed as the sum of squares of the differences between active power output, reactive power output, and planned values, while the risk cost function can be expressed as a weighted average of current utilization rate and risk indicators. The numerical form is configured by the operation and maintenance unit based on experience.

[0091] Regarding constraints, power constraints limit the active and reactive power output of the prefabricated substation to within the grid connection agreement and the rated capacity of the equipment; voltage constraints limit the bus voltage and critical node voltage to within the allowable operating range; current constraints ensure that the current in each circuit does not exceed the long-term and short-term allowable current values ​​of conductors, switchgear, and buses; and grounding safety constraints link the estimated grounding resistance and grounding current to the risk of grounding faults, preventing continued high-output operation under high-risk conditions. These constraints can be provided by constraint functions from a digital twin model. By performing power flow calculations and fault current calculations on different combinations of control variables, it can be determined whether the power, voltage, current, and grounding safety constraints are met.

[0092] The objective function and constraints constitute a finite-dimensional nonlinear optimization problem. The solution variables include the active power output setpoint, reactive power output setpoint, and control quantities corresponding to the relevant loop currents in the next control cycle. The control system can be solved using sequential quadratic programming or other mature constrained optimization algorithms to obtain an approximate optimal solution within a given computation time. The solution results are organized into a control quantity vector, which includes at least the active and reactive power reference values ​​of the grid-connected inverter, the switching states of capacitor banks and reactors, and the target voltage value in the voltage-reactive power control strategy.

[0093] After the control vector is issued, the field control device executes the corresponding power regulation and reactive power compensation operations, and coordinates and adjusts the action settings of the electrical protection devices based on the optimization results. For example, when the risk index is high and the current utilization rate of a certain circuit is close to the upper limit, the overcurrent protection settings and ground fault protection settings can be tightened within the allowable range to shorten the fault clearing time; when the risk index is low and the equipment margin is large, the protection settings can be appropriately relaxed to reduce unnecessary tripping risks. In terms of multi-frequency electrical excitation signal injection configuration, the control system dynamically selects the frequency combination and injection period that is more conducive to identifying key degradation parameters based on the latest risk index and parameter estimation results, so that subsequent state estimation and parameter estimation are more focused on high-risk parts, thus forming a closed-loop control process of "risk assessment - power regulation - parameter identification - risk update".

[0094] The risk-weighted objective function takes the predicted trajectories of active power output, reactive power output, and current as independent variables within a finite prediction time range. It sums the deviation costs of active power output and reactive power output from the power planning curve with the weighted costs of current on the risk index, and obtains the control vector corresponding to each prediction time through rolling optimization. The control vector corresponding to the current time is then executed in each control cycle.

[0095] In this embodiment, the risk-weighted objective function uniformly optimizes the predicted trajectories of active power output, reactive power output, and current of the new energy prefabricated substation within a finite prediction time range. At the beginning of each control cycle, the control system reads the current state from the power planning curve, the current operating condition, and the digital twin model, and sets a prediction time window containing several future control cycles. Within this prediction time window, assuming a series of reference trajectories for active power output and reactive power output to be optimized, as well as the resulting predicted current trajectories for each loop, the digital twin model calculates the corresponding bus voltage, current distribution, and risk index sequence time-by-time, given the control quantity trajectories.

[0096] To unify plan tracking and risk mitigation into a single optimization objective, this embodiment constructs a risk-weighted objective function in the following form:

[0097]

[0098] in, The total objective function value over the entire prediction time window. The number of prediction steps within the prediction time window. For the first The contribution of each predicted moment For the first Unproductive output at a predicted moment For the first The current vector at each predicted time point, For the first Risk indicators corresponding to each prediction time point This represents the weighting coefficient of risk cost relative to planning deviation cost. Planning deviation cost function. To measure the deviation of active and reactive power output from the power plan curve, it can be expressed as the sum of squares of the active and reactive power output deviations; risk cost function. This measure is used to assess how close the current is to the current limit at a given risk level. When the risk index is high and the current is close to the long-term allowable current or the short-term allowable current, the value of this measure increases significantly, thereby guiding the optimization results to reduce output or redistribute power flow.

[0099] While constructing the objective function, power constraints, voltage constraints, current constraints, and grounding safety constraints are applied at each prediction time. Power constraints ensure that active and reactive power outputs do not exceed the rated capacity of the prefabricated substation and its grid-connected inverters, and do not violate active and reactive power dispatch instructions issued by the upper-level power grid. Voltage constraints ensure that bus voltage and critical node voltage remain within allowable voltage ranges. Current constraints ensure that the current in each circuit does not exceed the long-term and short-term allowable currents of conductors, transformer windings, and switching equipment. Grounding safety constraints combine the aforementioned grounding-related parameter estimates with risk indicators to limit the maximum allowable output under high-risk conditions. All constraints are verified by the digital twin model at each prediction step through power flow calculations and fault current calculations.

[0100] The objective function and constraints combine to form a finite-dimensional constrained optimization problem, with the solution variables being the control vector corresponding to each moment within the prediction time window. The control vector includes at least the active power reference value and reactive power reference value of the grid-connected inverter, as well as the switching states of the reactive power compensation device and the voltage regulation device. The control system invokes a rolling optimization solver within each control cycle to numerically solve the objective function, obtaining a sequence of control vectors corresponding to each prediction moment within the entire prediction time window. To balance real-time performance and solution quality, sequential quadratic programming, interior-point methods, or other mature engineering numerical optimization methods can be used to obtain a feasible and approximately optimal control sequence within a preset computation time.

[0101] The rolling optimization execution strategy is "predict multiple steps, execute one step": within the current control cycle, only the control vector corresponding to the current moment is actually executed and sent to the field control device for adjusting the active and reactive power output of the grid-connected inverter, controlling the switching of the reactive power compensation device, and updating the injection configuration of the multi-frequency electrical excitation signal. When the next control cycle arrives, the latest parameter estimation vector, state estimation vector, operating condition vector, and risk index are re-acquired, the prediction time window and objective function are reconstructed, and a new control vector sequence is solved again. Through this rolling optimization mechanism, this embodiment can dynamically adjust the output and current levels according to changes in risk indicators while ensuring basic tracking of the power plan, so that the new energy prefabricated substation maintains a safety margin in long-term operation, while taking into account the output utilization rate and equipment life.

[0102] The control vector includes the active power reference value, reactive power reference value, and switching status of the capacitor bank and reactor connected to the grid-connected inverter of the new energy box-type substation. The control vector is sent to the grid-connected inverter and switching equipment through the field control device to realize power regulation and reactive power compensation.

[0103] In this embodiment, the control vector is generated by the upper-level control unit after solving the risk-weighted objective function, serving as the sole instruction carrier connecting the optimization calculation results and the field execution device. For each grid-connected inverter connected to the new energy prefabricated substation, the control vector includes at least the active power reference value and reactive power reference value of that inverter. For capacitor banks and reactors configured on the bus or feeder side of the prefabricated substation, the control vector also includes the switching status flags corresponding to each capacitor bank and each reactor. The switching status flags typically use binary values ​​to represent the on / off state or multi-valued values ​​to represent the tiered switching positions.

[0104] In system implementation, the station control layer or area controller expands the control vector obtained from rolling optimization according to the equipment list, forming a mapping table of "equipment identifier - control field". For grid-connected inverters, the active power reference value is written into the inverter's active power setpoint register, and the reactive power reference value or power factor setpoint value is written into the corresponding reactive power control register. After receiving the updated reference value, the field control device gradually adjusts the output through the inverter's internal current control loop and power control loop over several control cycles, so that the actual active and reactive power outputs track the target values ​​given in the control vector, while monitoring protection quantities such as voltage, current, and frequency to avoid over-limit operation. For capacitor banks and reactors, the corresponding switching status flags in the control vector are converted into specific switching outputs by the field control device. Through the action of the closing and opening mechanisms of medium-voltage or low-voltage switchgear, the capacitor banks and reactors are put into or taken out of service, thereby changing the reactive power compensation level and voltage support capability of the prefabricated substation.

[0105] To ensure the reliability and traceability of the control vector issuance process, the station control layer executes the process of "generating control vectors, encoding, sending, and reading back for confirmation" in a fixed sequence within each control cycle. During the encoding phase, the control vector is split into several messages according to the field communication protocol. Each message contains the target device address, control command type, and corresponding reference value or switching instruction. During the sending phase, the messages are sent to the corresponding circuit breaker control unit or inverter control unit via fieldbus or Ethernet. During the reading back phase, after the device completes its execution, the actual active power output, reactive power output, and switch status are read and compared with the target value in the control vector to determine whether the control was successfully executed and whether an alarm or degradation strategy needs to be activated. Through this method, this embodiment transforms the optimized active power reference value, reactive power reference value, and capacitor bank and reactor switching status into instructions that can be directly issued and executed in the existing automation system, realizing closed-loop control of power regulation and reactive power compensation in new energy prefabricated substations.

[0106] The operating settings of electrical protection devices include overcurrent protection settings and ground fault protection settings. When adjusting the operating settings of electrical protection devices, the amplitude is limited according to the preset upper and lower limits, and the maximum allowable change between adjacent control cycles is also limited.

[0107] In this embodiment, the electrical protection devices of the new energy prefabricated substation include overcurrent protection and ground fault protection configured on each feeder circuit and bus section. The protection setting calculation module periodically provides recommended overcurrent protection settings and ground fault protection settings based on the current distribution, ground current estimation results, and upper-level risk assessment results provided by the digital twin model. To avoid the impact of frequent and significant changes in protection settings on relay protection coordination and field operation, this embodiment adds a limiting and rate-of-change control link between the recommended settings and the actual issued settings, achieving gradual adjustment of the protection settings.

[0108] In practical implementation, the system pre-configures upper and lower limits for overcurrent protection settings and ground fault protection settings for each protection loop. The upper limit is determined based on the long-term allowable current of the primary equipment, conductor cross-section, and thermal stability capability, while the lower limit is determined based on the selective coordination principle and minimum sensitivity requirements. After obtaining the new recommended settings, the protection setting calculation module first checks whether the recommended settings exceed the corresponding upper or lower limit for each loop. If they do, the settings are truncated to the upper or lower limit, thus ensuring that the target settings generated at any given time do not exceed the pre-verified safety boundaries and avoiding severe protection mismatch due to algorithm anomalies or abnormal input data.

[0109] After the limit is applied, to control the slope of the setting change between adjacent control cycles, each loop is also configured with a maximum allowable change parameter to limit the adjustment range of the setting within a single control cycle. Specifically, at the beginning of the current control cycle, the system reads the overcurrent protection setting and ground fault protection setting actually executed in the previous control cycle and compares them with the target setting after the limit is applied this week. If the difference is less than or equal to the maximum allowable change, the target setting is directly adopted in this cycle; if the difference is greater than the maximum allowable change, only the maximum allowable change is adjusted along the direction of the target setting, and the adjusted value is used as the protection setting actually issued in this cycle. Through this gradual convergence method, the protection setting can be smoothly transitioned from the original level to the target level after risk optimization, without causing a leap in a short period of time. This preserves the original coordination relationship between upper and lower level protections and continuously incorporates risk assessment results.

[0110] The setting adjustment results are sent from the station control layer to each protection device via the protection communication protocol, correspondingly writing the current setting value for the overcurrent protection section, the current setting value for the ground fault protection section, or the zero-sequence current setting value. After the sending is completed, the system confirms whether the protection device has correctly received and applied the new setting value through the readback function or inspection message, and records the setting value change trajectory of each circuit in each control cycle in the operation log for subsequent review and operation and maintenance analysis. Through the above-mentioned limiting and rate of change control mechanism, this embodiment, while introducing online risk assessment-driven setting self-adaptation, ensures that the protection setting adjustment process is stable and controllable, and will not bring new uncertain risks to the protection safety and relay protection coordination of the new energy prefabricated substation.

[0111] In one specific embodiment, a photovoltaic new energy prefabricated substation with a grid connection capacity of 1MW is selected as the object. The low-voltage side of the prefabricated substation is a 0.4kV busbar, connected to two grid-connected inverters, each with a rated active power output of 500kW. Voltage and current acquisition devices and digital protection devices are installed on the busbar side, and the grounding grid resistance is approximately 0.5Ω during initial commissioning.

[0112] Under stable power output conditions at midday on a sunny day, the control system injects three discrete-frequency multi-frequency electrical excitation signals into the low-voltage bus, with frequencies of 210Hz, 260Hz, and 310Hz, and amplitudes approximately 1% of the rated voltage. Within one excitation cycle, the sampled effective phase voltage value of the 210Hz component is approximately 4.0V, the effective current value is approximately 0.40A, and the corresponding equivalent impedance magnitude is approximately 10Ω; the equivalent impedance of the 260Hz component is approximately 9.6Ω; and the equivalent impedance of the 310Hz component is approximately 9.3Ω. By synchronously acquiring the effective values ​​of active power, reactive power, and current, an operating condition vector is statistically obtained between adjacent excitation cycles. For example, in the current cycle, the active power is 0.94MW, the reactive power is -0.12Mvar, and the effective current value is approximately 1400A.

[0113] The initial parameters of the digital twin model were set according to the design values, such as a grounding branch resistance of 0.5Ω and a cable resistance of 0.08Ω / km. During the first 30 minutes of continuous operation, the joint state estimation and parameter estimation tracked the multi-frequency excitation response. In the parameter estimation vector, the grounding resistance component gradually converged from 0.50Ω to 0.72Ω, and the equivalent resistance of a certain outgoing cable converged from 0.16Ω to 0.21Ω. The bus voltage given by the state estimation vector was approximately 0.40kV, and the deviation of the current in each loop from the measured values ​​was controlled within 1%, indicating that the digital twin model has well fitted the current operating conditions.

[0114] Based on this, a digital twin model is used to predict the excitation response observation vector, which is then compared with the measured excitation response observation vector to obtain the residual signal. Taking the 260Hz component as an example, the model predicts an effective current value of 0.36A, the measured value is 0.40A, and the residual is 0.04A, showing a continuous increasing trend within the most recent 60-minute time window. The residual characteristics are combined with features such as the increasing trend of grounding resistance, cable resistance, and operating current growth to form a comprehensive feature vector, which serves as the input for fault diagnosis and life assessment.

[0115] The fault mode classification model trained based on historical samples outputs probabilities for three fault modes under the current comprehensive feature vector: grounding degradation fault mode probability is approximately 0.60, joint overheating fault mode probability is approximately 0.25, and insulation aging fault mode probability is approximately 0.15. The lifetime prediction model, under the same input, provides corresponding remaining lifetime estimates: grounding degradation fault mode remaining lifetime is approximately 12,000 hours, joint overheating fault mode remaining lifetime is approximately 16,000 hours, and insulation aging fault mode remaining lifetime is approximately 22,000 hours. To quantify the current overall operational risk, the following risk index calculation formula is used:

[0116]

[0117] in, As a risk indicator, For the first The weights of each fault mode are as follows: grounding degradation fault mode has a weight of 1.0, joint overheating fault mode has a weight of 0.7, and insulation aging fault mode has a weight of 0.9. For the first The probability value of each failure mode; For the first The remaining useful life of each failure mode; To map the remaining useful life to a dimensionless risk factor function, we take [the following] here. , The reference lifespan is 20,000 hours. Substituting the above values, the risk index is approximately 1.35, which is higher than the preset risk threshold of 1.0, indicating that the current grounding degradation risk is relatively high.

[0118] Under risk constraints, a risk-weighted objective function was constructed to perform rolling optimization on the predicted trajectories of active power output, reactive power output, and current at six control moments within the next 30 minutes. Without control measures, the power planning curve requires the prefabricated substation to maintain 0.95MW of active power output during this period. The digital twin model predicts that under high summer temperatures, the peak current of a certain outgoing line will approach 95% of the allowable current limit, while the grounding current will approach 80% of the protection setting under slight imbalance conditions. The optimized solution suggests reducing the total active power output to 0.85MW, adjusting the reactive power operating point from -0.12Mvar to -0.05Mvar, and simultaneously disconnecting one set of terminal capacitor banks, causing the bus voltage to slightly decrease from 1.05 per unit to 1.02 per unit; correspondingly, the maximum current of each circuit will decrease to approximately 80% of the allowable current limit, and the grounding current will decrease to approximately 70% of its original value.

[0119] After executing the aforementioned control vector, the station control layer simultaneously fine-tunes the overcurrent protection and ground fault protection action settings. For example, if the overcurrent protection setting for a certain outgoing line was 600A in the previous cycle, with an upper limit of 650A, a lower limit of 450A, and a maximum allowable variation of 20A, the recommended setting for this cycle, based on the new thermal stability margin, is 560A. After amplitude limiting confirmation that it falls within the upper and lower limits, the actual issued setting is adjusted from 600A to 580A. The ground fault protection setting is adjusted from 80A to 75A in a similar manner. After 2-3 control cycles, the settings gradually converge to the new optimal level.

[0120] One hour after the control strategy was executed, state and parameter estimations were performed again. The estimated grounding resistance remained at approximately 0.75Ω, but due to the reduction in current level and voltage offset, the residual signal decreased significantly. Taking the 260Hz component as an example, the residual decreased from 0.04A to 0.02A. The fault mode probabilities corresponding to the new comprehensive feature vector became: grounding degradation fault mode probability approximately 0.40, joint overheating fault mode probability approximately 0.30, and insulation aging fault mode probability approximately 0.30. The corresponding remaining service lives given by the life prediction model increased to approximately 16,000h, 18,000h, and 23,000h, respectively. Substituting into the same risk index calculation formula, the risk index decreased to approximately 0.98, below the risk threshold. This indicates that by jointly adjusting active power output, reactive power compensation, and protection settings, while ensuring that the planned output is basically met, the overall operational risk of the prefabricated substation was successfully reduced and the degradation process of key components was slowed down.

[0121] This embodiment demonstrates that the present invention does not simply rely on a single measurement for threshold control. Instead, it connects "clear visibility" and "reasonable adjustment" through multi-frequency electrical excitation, digital twin state estimation, degradation pattern recognition, and risk quantification. Quantitative results show that when grounding resistance has significantly increased, traditional setting and output strategies that only consider the upper limit of current are insufficient to reflect risks in a timely manner. However, the present invention, through risk-indicator-driven rolling optimization, reduces current utilization from 95% to around 80%, and the risk index from 1.35 to 0.98. This provides ample time for subsequent planned maintenance without requiring immediate shutdown, and offers more refined technical means for the safe operation of prefabricated substations in scenarios with a high proportion of renewable energy integration.

[0122] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for estimating and controlling the state of a new energy transformer substation, characterized in that, Includes the following steps: Injecting multi-frequency electrical excitation signals into the new energy prefabricated substation and simultaneously acquiring voltage and current signals, the excitation response observation vector and operating condition vector are obtained. A digital twin model is established based on the primary wiring topology and the grounding grid topology. Joint state estimation and parameter estimation are performed using the excitation response observation vector and the operating condition vector to obtain the parameter estimation vector and the state estimation vector. A comprehensive feature vector representing the degradation state is constructed using parameter estimation vector, state estimation vector, and operating condition vector. The comprehensive feature vector is then input into the fault diagnosis and life assessment model to obtain risk indicators. The construction of a comprehensive feature vector representing the degradation state using parameter estimation vector, state estimation vector, and operating condition vector includes: calculating the predicted excitation response observation vector using a digital twin model; constructing a residual signal using the difference between the excitation response observation vector and the predicted excitation response observation vector; and concatenating the residual signal with the change characteristics of the parameter estimation vector, state estimation vector, and operating condition vector within a preset time window to obtain the comprehensive feature vector. Based on risk indicators, a risk-weighted objective function with active power output, reactive power output, and current as independent variables is constructed. Under the conditions of satisfying power constraints, voltage constraints, current constraints, and grounding safety constraints, the control vector is obtained. The control vector is used to adjust the power output and the action setpoint of the electrical protection device of the new energy box-type substation, and to update the injection configuration of the multi-frequency electrical excitation signal. The risk-weighted objective function uses the predicted trajectories of active power output, reactive power output, and current as independent variables within a finite prediction time range. It sums the deviation costs of active power output and reactive power output from the power planning curve with the weighted costs of current on the risk index, and obtains the control vector corresponding to each prediction time through rolling optimization. In each control cycle, the control vector corresponding to the current time is executed.

2. The method according to claim 1, characterized in that, The multi-frequency electrical excitation signal is obtained by superimposing a set of discrete frequency sinusoidal electrical excitation signals. The voltage and current signals are sampled synchronously under a unified time base. The amplitude and phase of the voltage and current signals at each discrete frequency are organized into an excitation response observation vector through frequency domain analysis. The operating condition vector includes the active power, reactive power and current RMS values ​​obtained statistically between adjacent excitation cycles.

3. The method according to claim 1, characterized in that, The digital twin model represents the primary equipment terminals and grounding points of the new energy prefabricated substation as graph nodes, and the conductors, transformer windings and grounding leads as graph edges. Resistance parameters, inductance parameters, capacitance parameters and grounding resistance parameters are set on each graph edge. The digital twin model uses the resistance parameters, inductance parameters, capacitance parameters and grounding resistance parameters in the parameter estimation vector as the electrical parameters of each graph edge, the voltage variables in the state estimation vector as the graph node voltages, and the current variables in the state estimation vector as the graph edge currents.

4. The method according to claim 1, characterized in that, The joint state estimation and parameter estimation adopt the extended Kalman filter algorithm and the recursive least squares algorithm. The extended Kalman filter algorithm uses the excitation response observation vector and the operating condition vector as the observation to update the state estimation vector, while the recursive least squares algorithm uses the state estimation vector as the regression variable and the excitation response observation vector as the observation to update the parameter estimation vector.

5. The method according to claim 1, characterized in that, The fault diagnosis and life assessment model includes a fault mode classification model and a life prediction model. The fault mode classification model takes the comprehensive feature vector as input and outputs the probability values ​​of grounding degradation fault mode, joint overheating fault mode and insulation aging fault mode. The life prediction model takes the comprehensive feature vector and fault mode probability values ​​as input and outputs the remaining service life of each fault mode.

6. The method according to claim 5, characterized in that, The risk index is a scalar risk measure obtained by weighting the failure mode probability value output by the failure mode classification model and the remaining useful life of the corresponding failure mode output by the lifetime prediction model according to preset weights.

7. The method according to claim 1, characterized in that, The control vector includes the active power reference value, reactive power reference value, and switching status of the capacitor bank and reactor connected to the grid-connected inverter of the new energy box-type substation. The control vector is sent to the grid-connected inverter and switching equipment through the field control device to realize power regulation and reactive power compensation.

8. The method according to claim 1, characterized in that, The operating settings of electrical protection devices include overcurrent protection settings and ground fault protection settings. When adjusting the operating settings of electrical protection devices, the amplitude is limited according to the preset upper and lower limits, and the maximum allowable change between adjacent control cycles is also limited.

Images