Novel electric power system generalized load component online detection and state tracking method
By employing interactive multi-model algorithms and statistical clustering methods in novel power systems, an offline sub-model set including distributed power sources and energy storage systems is established. This solves the problem of dynamic time-varying characteristics in load modeling of novel power systems, enabling real-time and accurate detection and status tracking of generalized loads, and supporting the safe and stable operation of the system.
Patent Information
- Application Number
- CN202511426735.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-30
- Publication Date
- 2026-01-09
AI Technical Summary
Existing power system load modeling methods cannot accurately reflect the dynamic time-varying characteristics of new power systems, especially in the case of distributed power sources and energy storage integration, where there are problems such as insufficient model representativeness and difficulty in real-time detection.
An offline sub-model set including distributed power sources and battery energy storage systems is established by combining the interactive multi-model algorithm (IMM) with statistical clustering. Generalized load modeling is performed using online data, and model filtering and probability updates are performed using an unscented Kalman filter to achieve real-time detection and tracking of the generalized load status.
It enables real-time and accurate detection and status tracking of generalized loads in new power systems, can identify anomalies and faults, provides key data support for the safe and stable operation of the system, and has high model fitting accuracy and fast iteration speed.
Smart Images

Figure BDA0005624625010000031 
Figure BDA0005624625010000032 
Figure BDA0005624625010000041
Abstract
Description
Technical Field
[0001] This invention relates to a novel online detection and status tracking method for generalized load components in power systems, belonging to the field of power system operation monitoring technology. Background Technology
[0002] Power system load modeling is a crucial technology widely studied in system planning, equipment control, and dispatching operations. It obtains reliable quantitative analysis results by accurately reflecting the external characteristics of system components. With the increasing integration of distributed power sources and energy storage, new power systems exhibit characteristics such as changes in power structure, increased uncertainty, structural complexity, and supply-demand ambiguity. As traditional power systems transform into new power systems, traditional loads are transformed into generalized loads. Traditional modeling methods and model structures are increasingly unable to accurately describe the characteristics of new power systems, necessitating the establishment of a generalized load model that can accurately capture the dynamic characteristics of new power systems.
[0003] For modeling the generalized load characteristics in new power systems, Wu Feng et al. combined photovoltaic systems with traditional loads based on the characteristics of similar loads in photovoltaic power generation systems, reducing model complexity and identification difficulty. However, when the system experiences near-port faults, the voltage drop is too large to effectively describe the load. Qu Xing et al. proposed a load-oriented second-order equivalent model of PV, consisting of a series RLC circuit connected in parallel with a constant power source, specifically for the transient process of PV power generation. Qu Xing et al., through simulation analysis of the transient operating characteristics of BESS (Boundary Energy Sequence Flow) systems, proposed a load-oriented equivalent model of BESS, further establishing a generalized integrated load model of the distribution network considering BESS. The above literature mainly focuses on the study of single generalized loads in new power systems, providing model references for online modeling of generalized load characteristics in new power systems. However, since it does not describe the overall generalized load characteristics at the macroscopic level of the system, it has significant limitations in practical applications.
[0004] Power systems contain a massive number of load components. Describing each type of component using an equivalent model would be incredibly complex. Scholars have conducted system-level modeling studies on power systems, employing static ZIP models (constant impedance, constant power, constant current) to characterize the overall power system. However, these models cannot represent the dynamic characteristics of the system. He et al. introduced dynamic models into load models, forming a classic load model (CLM) combining a third-order induction motor (IM) and ZIP. Tang et al. further considered the impact of power system network structure on load modeling, proposing the SLM model. However, all of the above literature uses offline model construction methods, resulting in relatively fixed model structures that cannot accurately reflect the dynamic and time-varying characteristics of new power systems.
[0005] Traditional modeling research often employs offline modeling methods, resulting in models that are almost static. However, power systems, composed of numerous complex load components, generate massive amounts of load characteristic data during operation, and their load composition may also change. Offline modeling cannot reflect the online dynamic characteristics of the system and carries the risk of insufficient model representativeness. Therefore, scholars have proposed the concept of online modeling. Yinyin Ge et al. proposed an online load modeling method for microgrids based on the real-time nature of online monitoring information from phasor measurement units (PMUs). Cai Shuping et al., relying on the big data environment, proposed an online modeling method based on online support vector regression and Fisher information meteorological factor processing. Zhu Jianquan et al. introduced the interactive multi-model algorithm from the field of maneuver tracking into the field of load modeling, proposing an online load modeling method based on the IMM algorithm. Zhongqiang Li et al., building on the work of Zhu Jianquan et al., further introduced a model selection mechanism to select the optimal sub-model set. However, their sub-model sets all use the same model structure, and the load research objects are subjectively selected, posing a risk to the representativeness of the system, and failing to consider the access of generalized loads such as distributed power sources.
[0006] While the above online modeling methods can reflect the dynamic characteristics of the system to some extent, they still employ a single model structure, namely a fixed model set of ZIP plus IM to describe all loads in the system. When applied to new power systems with numerous generalized loads, this poses a risk of insufficient model representativeness. The IMM algorithm transforms the load dynamic modeling problem into the problem of accurately tracking time-varying load characteristics, providing a feasible approach to solving the time-varying problem of generalized load dynamic modeling in integrated energy power systems. However, existing online modeling research based on the IMM algorithm is mostly aimed at traditional power systems and uses the same offline sub-model structure for interaction, making it difficult to meet the online modeling needs of new power systems. Moreover, existing modeling methods often remain focused on the "model" itself, failing to effectively transform the results of online modeling into real-time, dynamic detection information of the system's generalized load status. They cannot perceive changes in load composition in real time, nor can they track the intrinsic operating status of distributed power sources and energy storage, making it difficult for dispatch centers to obtain this crucial detection data for real-time control and decision-making. This invention addresses the issue of the simplistic structure of offline sub-models in the IMM algorithm. It expands the offline sub-models of the IMM algorithm by employing different types of offline sub-models to describe various generalized loads existing in novel power systems, thereby enhancing the rationality and accuracy of the overall model construction. Furthermore, to address the ambiguity in sub-model load selection in existing interactive multi-model modeling, this invention utilizes statistical clustering to refine the construction of generalized load sub-models. Combined with sub-model structure expansion, it proposes an online detection method for generalized load components in novel power systems, specifically comprising two parts: establishing an offline sub-model set and online information interaction modeling. Summary of the Invention
[0007] To address the problems existing in the prior art, this invention provides a novel online detection and state tracking method for generalized load components in power systems. It aims to solve the technical challenge of existing methods failing to perform real-time and accurate state detection and tracking of generalized loads. This invention not only boasts high model fitting accuracy and fast iteration speed, but also enables real-time and accurate identification and early warning of generalized load anomalies, faults, or operational deviations, providing crucial data support for the safe and stable operation of the system.
[0008] The technical solution of this invention is: a novel online detection and state tracking method for generalized load components in a power system, the method comprising:
[0009] Step 1: Select generalized load cells based on optimized statistical clustering method;
[0010] Step 2: Establish and expand the offline sub-model set that includes distributed power sources and battery energy storage systems, and discretize the expanded offline sub-model set.
[0011] Step 3: Based on the interactive multi-model algorithm, perform online modeling of generalized load using online power system data;
[0012] Furthermore, the method also includes:
[0013] Step 4: Based on the established generalized load online model, perform real-time detection of the generalized load composition of the new power system, dynamic tracking of total load power, and online identification of key internal state variables.
[0014] Further, Step 1 includes:
[0015] Step 1.1: Construct the characteristic parameter matrix;
[0016] Taking m equipment components in the power system as research objects, each object is represented by n characteristic data, thus establishing the characteristic parameter matrix U*, as shown in equation (1);
[0017]
[0018] Where, x mn This represents the nth feature data of the mth device component;
[0019] The characteristic parameter matrix U* is formatted as shown in equation (2);
[0020]
[0021] In the formula, x′ mn x represents the formatted values of each element in U*.n This represents the nth feature data.
[0022] The data is normalized to between 0 and 1 to eliminate the influence of dimensions; the fuzzy similarity relationship of the equipment components is determined by multivariate analysis, a similarity matrix R is established, and the transitive closure t(R1) of the similarity matrix is obtained by squaring, so that it satisfies the relationship of equation (3);
[0023] R1→R1 2 →R1 4 →...(3) After performing a finite number of K / 2 operations on the above model, the transitive closure can be determined, as shown in equation (4);
[0024] R1 K =t(R1)=R1 K ·R1 K (4)
[0025] Step 1.2: Construct the domain of discourse for the basic unit device:
[0026] Based on the analysis and research of the basic unit operating conditions during normal system operation, according to industry classification, several typical "users" that can reflect the production characteristics of each industry are selected for preliminary investigation based on practical experience; on the basis of the preliminary selection results, the generalized load units of "users" that are not typical in reality are removed by using the mathematical method of fuzzy clustering, and the generalized load characteristics are further integrated; the percentage of equipment parameters of each basic unit is calculated based on the statistical physical characteristic parameters of equipment components, and the universe of discourse of the basic unit equipment is established, as shown in Equation (5):
[0027]
[0028] The same maximum value normalization method as Equation (2) is used to format each column of the domain L1;
[0029] Step 1.3: Construct a fuzzy similarity matrix;
[0030] The similarity coefficient r is introduced using the minimax method. mn , representing the m-th and n-th feature data of two samples, x m With x n To determine the degree of similarity between them, a fuzzy similarity matrix R1 is established, as shown in equation (6):
[0031]
[0032] The maximum-minimum method is shown in equation (7):
[0033]
[0034] In the formula, x nj,x mj represent the nth column element and the mth row element in U*, respectively; ∧ and ∨ represent the intersection and union operations, respectively.
[0035] Step 1.4: Construct an equivalent matrix that reflects the physical characteristics of each type of basic unit equipment;
[0036] The transitive closure t(R1) of the similarity matrix is solved using equations (3)-(4), and the equivalent matrix reflecting the physical characteristics of each type of basic unit equipment is obtained, as shown in equation (8):
[0037]
[0038] By selecting an appropriate value of λ to construct the cutoff matrix, the fuzzy clustering results are obtained. When λ = 0.4, the cutoff matrix t(0.4) is shown in equation (9):
[0039]
[0040] Furthermore, Step 2 includes:
[0041] Step 2.1: Construct a generalized load mathematical model;
[0042] Step 2.1.1: Construct an equivalent BESS model;
[0043] The BESS first-order RC equivalent circuit model is adopted, and the BESS first-order RC equivalent circuit model is shown in Equation (10).
[0044]
[0045] In the formula: U BESS (t), U oc (t) represent the port voltage and open-circuit voltage of the battery model at time t, respectively; R p U p and C p These are the battery's polarization internal impedance, voltage, and capacitance, respectively; R0 is the battery's equivalent internal resistance.
[0046] The state-of-charge expression of BESS obtained by the ampere-hour integration method is shown in the figure below;
[0047]
[0048] In the formula: ξ SOC (t) represents the state of charge of BESS at time t; Q N η is the rated capacity of the battery; η is the coulomb efficiency coefficient; I1 is the equivalent current in the first-order RC equivalent circuit of BESS;
[0049] Step 2.1.2: Construct a PV equivalent model;
[0050] The P-RLC model is selected as the PV equivalent sub-model for discretization, and the power factor of the PV power generation system is close to 1, and its reactive power characteristics are not considered; the PV equivalent sub-model is shown in Equation (12)-Equation (13).
[0051]
[0052] P PV =P0+I2|U R | (13)
[0053] In the formula: I2 is the current in the dynamic partial loop of the PV model; P PV U represents the overall output power of the model; R, L, and C represent the equivalent resistance, inductance, and capacitance values of the dynamic part of the PV model, respectively; U C For the dynamic portion of the capacitor voltage, U PV P0 is the grid terminal voltage, and P0 is the static active power output fixed value.
[0054] Step 2.2: Discretization of the equivalent model;
[0055] Step 2.2.1: Construct the state equations and measurement equations of the sub-model;
[0056] When using the interactive multi-model algorithm for online modeling, the offline sub-model needs to be discretized; the state equations and measurement equations of the discretized sub-model are shown in equations (14)-(15);
[0057] x i (k)=F i (k-1)x i (k-1)+ω i (k-1) (14)
[0058] z i (k)=H i (k)x i (k)+v i (k) (15)
[0059] In the formula: x i (k), z i (k) represents the state vector and measurement vector of the i-th sub-model at time k; F i (k-1) is the state transition matrix of the i-th sub-model at time k-1; H i (k) is the measurement matrix of the i-th sub-model at time k; ω i (k-1) and v i (k) are the Gaussian white noise matrices of the i-th sub-model at times k-1 and k, respectively, with a mean of 0;
[0060] Step 2.2.2: Construct the discretization of IM;
[0061] The IM in the model adopts the third-order IM model, and its state equation is shown in equation (16);
[0062]
[0063] In the formula: E′ d E′ q The transient potentials on the d and q axes are respectively; X and X′ are the synchronous reactance and transient reactance, respectively; T′0 is the transient open-circuit time constant; w N ,w m These are the rated speed and rotor speed, respectively; s is the slip rate; i d i q These are the transient current components on the d and q axes, respectively; T e ,T m These are the electromagnetic torque and mechanical torque of the rotor, respectively; H is the inertial time constant.
[0064] The measurement equations for the third-order induction motor model are shown in equation (17);
[0065]
[0066] in,
[0067]
[0068] In the formula: P m Q m These represent the active and reactive power of IM, respectively; U is the voltage value of the load node; R s For IM stator resistance;
[0069] Step 2.2.3: Discretization of the BESS equivalent model;
[0070] Discretize equations (10) and (11) according to the forms of equations (14) and (15) to obtain the state equation of the BESS equivalent model as shown in equation (24) and the measurement equation as shown in equation (25). The specific derivation process is as follows:
[0071]
[0072] In the formula, T is the unit time interval;
[0073] The state equations of the BESS equivalent model are thus obtained, as shown in the following equation:
[0074]
[0075] Since the values of each term in matrix Y approach zero, they are equivalent to noise terms. Finally, the simplified state equations and measurement equations of the BESS equivalent model are as follows:
[0076]
[0077] In the formula: P ES (k) represents the output power of BESS at time k, ω BESS v ES Both are Gaussian white noise in the state equations, U ES (k) represents the output voltage, and Z represents the parallel impedance;
[0078] Step 2.2.4: Discretization of the PV equivalent model;
[0079] By deriving equations (12) and (13), the dynamic state equation and measurement equation of the PV equivalent model are obtained as shown in equations (33) and (34). The specific derivation process is as follows:
[0080]
[0081] In the formula, U R U L These represent the resistor voltage and capacitor voltage in the generator dynamic model, respectively.
[0082] In steady state, the dynamic partial capacitor voltage U C Equal to the grid terminal voltage U PV The internal current is zero.
[0083]
[0084] In the formula, U′ C (0) represents the transient port voltage; u C (t) represents the capacitor port voltage at time t; I2(t) represents the dynamic partial loop current at time t; I2(0) represents the dynamic partial loop current at the initial time; I2(k+1) represents the dynamic partial loop current at time k+1; f = 50H Z ;s 1,2 Let s1 and s2 be as shown in Equation 31; U C (k) represents the terminal voltage at time k; K1 is a constant coefficient obtained from solving the differential equation; v PV ω PV All are Gaussian white noise; The initial voltage at the capacitor port; The current in the dynamic partial loop is at the initial moment.
[0085] Furthermore, Step 3 includes:
[0086] Step 3.1, Input Interaction;
[0087] As input for generalized load modeling, online power system data specifically includes the voltage amplitude U of load nodes. (k) The system frequency f, the total active power p injected into the node, the reactive power Q, the impedance value z, the state vectors of each sub-model, and Gaussian white noise, etc., need to be preprocessed before input interaction. This includes data cleaning to remove outliers and handle communication interruptions, as well as suppressing measurement noise through digital filtering technology, and finally forming a clean and aligned real-time measurement sequence Z(k), which serves as the direct input for subsequent model filtering and state updates.
[0088] At time k-1, each sub-model corresponds to an optimal state vector x. i (k-1) and estimated covariance matrix P i (k-1), and then the state vectors and covariance matrices of each sub-model are weighted and fused according to the correctness probability obtained from the Markov operation matrix to obtain the fused optimal state vector x. o (k-1) and covariance matrix P o (k-1):
[0089]
[0090] In the formula, λ IM (k-1), λ ZIP (k-1), λ PV (k-1), λ BESS (k-1), λ i (k-1) is the state transition matrix of the i-th sub-model at time k-1; x IM (k-1), x ZIP (k-1), x PV (k-1), x BESS (k-1) represents the state vector of the i-th sub-model at time k-1; n is the number of generalized load types, including energy storage, photovoltaic, and generator.
[0091] Step 3.2, Model Filtering;
[0092] The fused x o (k-1) and P o (k-1) is used as the input variable, and an unscented Kalman filter is used for filtering and updating. The predicted states of each sub-model are then determined. and predicted covariance As shown in equations (38)-(39);
[0093]
[0094] in:
[0095]
[0096] In the formula, K i (k) represents the filter gain; s i (k) represents the residual covariance; Z(k) represents the fitted actual value; r i (k) represents the standard deviation of the measurement noise; Predict the state and prediction covariance for each sub-model; F represents the predicted state of the fused overall model. i Markov matrix coefficients; Q i A randomly generated n*n standard normal distribution random number matrix with mean 0 and variance 1; H i (k) is the measurement matrix of the i-th sub-model at time k;
[0097] Step 3.3: Probability Update;
[0098] The maximum likelihood function method is adopted to calculate the weight of each sub-model by the similarity between the current model and the target state. The maximum likelihood function is calculated as shown in Equation (41).
[0099]
[0100] The weight calculation formulas for each sub-model after the update are shown in equation (42);
[0101]
[0102] Step 3.4, Interactive Output;
[0103] The updated state is fused and estimated based on the latest weights of each sub-model to obtain the final output result;
[0104]
[0105] Furthermore, Step 4 includes:
[0106] Step 4.1, Dynamic tracking and detection of total system load power: Based on the weights u of each sub-model at time k obtained from the interactive output in Step 3. i (k) and its measurement output Z i (k), through weighted fusion, calculates and outputs the tracking and detection values of active power and reactive power of the generalized load of the system in real time;
[0107] Step 4.2, Online Identification and Detection of Generalized Load Composition Ratio: Directly record and output the weight probability set {u1(k), u2(k), u3(k)…u} of each sub-model at time k obtained through probability update in Step 3. n(k)}, and use this set as the detection result of the real-time composition ratio of various generalized loads in the system;
[0108] Step 4.3, Online identification and detection of key internal state variables: Extract and output the state of charge ξ of the battery energy storage system sub-model from the optimal state vector x0(k) obtained after the interactive output in Step 3. SOC (t) and the detection output state components of the induction motor sub-model, the state of charge of the energy storage sub-model and the expression of the detection output state components of the induction motor sub-model are as follows:
[0109]
[0110]
[0111] The present invention also provides a novel online detection and state tracking system for generalized load components of a power system, the system comprising: a module for executing the aforementioned novel online detection and state tracking method for generalized load components of a power system.
[0112] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the novel online detection and state tracking method for generalized load components in a power system.
[0113] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the novel online detection and state tracking method for generalized load components in a power system.
[0114] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the novel online detection and state tracking method for generalized load components in a power system.
[0115] The beneficial effects of this invention are:
[0116] (1) This invention uses statistical clustering to select the massive generalized loads in the new power system, which greatly reduces the complexity of modeling while ensuring the representativeness of the overall system, provides support for the subsequent establishment of IMM offline sub-models, and solves the ambiguity problem in the selection of offline sub-model sets when using IMM for load modeling.
[0117] (2) During the establishment of the IMM offline sub-model, the present invention classifies the selected generalized loads into four categories based on their own characteristics: static generalized loads, rotating generalized loads, BESS generalized loads and PV generalized loads, and performs discretization processing, thereby expanding the interactive multi-model offline sub-model set and optimizing the problem of insufficient representativeness of the single structure model for generalized loads and new power systems.
[0118] (3) The new power system generalized load online model based on IMM established by this invention has a fast calculation speed and better overall output fitting accuracy compared with the new power system generalized load model with the same sub-model. When the system load components and system structure change, the proposed model has a good tracking effect and can accurately reflect the dynamic characteristics of the new power system, which meets the requirements of online modeling.
[0119] (4) The process and method for establishing a new power system generalized load online model based on IMM proposed in this invention have strong universality, but the composition of the internal offline sub-model set and the sub-model parameters have certain regional characteristics. When using it, it should be modified accordingly based on the actual situation. The proposed model has good generalization within a voltage drop of 30%, and its error value after relative stability is less than 5%. Moreover, the interpolation characteristics are better than the extrapolation characteristics, that is, the model has good descriptive ability within the applicable range.
[0120] (5) The generalized load online model constructed in this invention aims to achieve online detection of the generalized load components of the power system. Through this model, the total load power of the system, the weights of each sub-model (i.e., the load composition ratio), and key state variables can be tracked and identified in real time. This breaks through the limitations of traditional offline modeling or single-model online modeling, which can only obtain a fixed model. The method ultimately focuses on "online detection and state tracking," providing visualization means and decision-making basis for real-time scheduling, safety early warning, and optimized operation of the power grid. Attached Figure Description
[0121] Figure 1 This is a three-dimensional schematic diagram of dynamic clustering of typical equipment in a novel power system generalized load according to the present invention;
[0122] Figure 2 This is a schematic diagram of the structure of the novel generalized load model for power systems in this invention;
[0123] Figure 3 This is a schematic diagram comparing the active power output in scenario 1 of the novel online modeling of generalized loads in the power system of this invention.
[0124] Figure 4 This is a schematic diagram comparing the active power output in scenario 2 of the novel online modeling of generalized loads in the power system of this invention.
[0125] Figure 5 This is a schematic diagram illustrating the reactive power generalization error in the novel online modeling of generalized loads in the power system of this invention.
[0126] Figure 6 This is a schematic diagram of the dynamic model of a PV generator in the novel online modeling of generalized loads in a power system according to the present invention.
[0127] Figure 7 This is a schematic diagram illustrating the principle of the novel online generalized load modeling (IMM) algorithm for power systems in this invention.
[0128] Figure 8 This is a schematic diagram of the active power error of the IEEE-33 node distribution network with a 40% voltage drop in this invention;
[0129] Figure 9 This is a schematic diagram illustrating the error in the study of the active power generalization of the 30% voltage drop in the distribution network at the EEE-33 node in this invention. Detailed Implementation
[0130] Example 1: As Figures 1-9 As shown, a novel online detection and state tracking method for generalized load components in a power system is proposed, the method comprising:
[0131] Step 1: Select generalized load cells based on optimized statistical clustering method;
[0132] Step 2: Establish and expand the offline sub-model set that includes distributed power sources and battery energy storage systems, and discretize the expanded offline sub-model set.
[0133] Step 3: Based on the interactive multi-model algorithm, perform online modeling of generalized load using online power system data;
[0134] Step 4: Based on the established generalized load online model, perform real-time detection of the generalized load composition of the new power system, dynamic tracking of total load power, and online identification of key internal state variables.
[0135] Further, Step 1 includes:
[0136] Step 1.1: Construct the characteristic parameter matrix;
[0137] Taking m equipment components in the power system as research objects, each object is represented by n characteristic data, thus establishing the characteristic parameter matrix U*, as shown in equation (1);
[0138]
[0139] Where, x mn This represents the nth feature data of the mth device component;
[0140] Since the characteristic data dimensions and orders of magnitude of the equipment components constituting the classification object are not necessarily the same, it is necessary to format the characteristic parameter matrix U*, as shown in equation (2).
[0141]
[0142] In the formula, x′ mn x represents the formatted values of each element in U*. n This represents the nth feature data.
[0143] The data is normalized to between 0 and 1 to eliminate the influence of dimensions; the fuzzy similarity relationship of the equipment components is determined by multivariate analysis, a similarity matrix R is established, and the transitive closure t(R1) of the similarity matrix is obtained by squaring, so that it satisfies the relationship of equation (3);
[0144] R1→R1 2 →R1 4 →...(3) After performing a finite number of K / 2 operations on the above model, the transitive closure can be determined, as shown in equation (4);
[0145] R1 K =t(R1)=R1 K ·R1 K (4)
[0146] Step 1.2: Construct the domain of discourse for the basic unit device:
[0147] Taking a new type of power system in a certain region as the research object, this study analyzes the basic unit operating conditions under normal system operation. According to industry classification, based on practical experience, several typical "users" that can reflect the production characteristics of each industry are selected for preliminary investigation. Based on the preliminary selection results, the generalized load units of "users" that are not typical in reality are removed by using the mathematical method of fuzzy clustering, and the generalized load characteristics are further integrated. The percentage of equipment parameters of each basic unit is calculated based on the statistical physical characteristic parameters of the equipment components, and the domain of discourse of the basic unit equipment is established, as shown in Equation (5).
[0148]
[0149] Since the characteristic data of each power load are quite different, the same maximum value normalization method of formula (2) is used to format each column of the domain L1 to eliminate the influence of the dimension.
[0150] Step 1.3: Construct a fuzzy similarity matrix;
[0151] To establish the fuzzy similarity matrix, the minimax method is used to introduce the similarity coefficient r. mn, representing the m-th and n-th feature data of two samples, x m With x n To determine the degree of similarity between them, a fuzzy similarity matrix R1 is established, as shown in equation (6):
[0152]
[0153] The maximum-minimum method is shown in equation (7):
[0154]
[0155] In the formula, x nj ,x mj represent the nth column element and the mth row element in U*, respectively; ∧ and ∨ represent the intersection and union operations, respectively.
[0156] Step 1.4: Construct an equivalent matrix that reflects the physical characteristics of each type of basic unit equipment;
[0157] The transitive closure t(R1) of the similarity matrix is solved using equations (3)-(4), and the equivalent matrix reflecting the physical characteristics of each type of basic unit equipment is obtained, as shown in equation (8):
[0158]
[0159] By selecting an appropriate value of λ to construct the cutoff matrix, the fuzzy clustering results are obtained. When λ = 0.4, the cutoff matrix t(0.4) is shown in equation (9):
[0160]
[0161] Furthermore, Step 2 includes:
[0162] Step 2.1: Construct a generalized load mathematical model;
[0163] Step 2.1.1: Construct an equivalent BESS model;
[0164] Common BESS equivalent models include electrochemical models and equivalent circuit models. Among them, electrochemical models are mostly used to study the influence of internal materials on battery performance; equivalent circuit models use basic physical components to simulate the external electrical characteristics of BESS during normal operation, and their structure is simple and their physical meaning is clear. The first-order RC equivalent circuit model of BESS is adopted, and the first-order RC equivalent circuit model of BESS is shown in Equation (10);
[0165]
[0166] In the formula: U BESS (t), U oc (t) represent the port voltage and open-circuit voltage of the battery model at time t, respectively; Rp U p and C p These are the battery's polarization internal impedance, voltage, and capacitance, respectively; R0 is the battery's equivalent internal resistance.
[0167] The state-of-charge expression of BESS obtained by the ampere-hour integration method is shown in the figure below;
[0168]
[0169] In the formula: ξ SOC (t) represents the state of charge of BESS at time t; Q N η is the rated capacity of the battery; η is the coulomb efficiency coefficient; I1 is the equivalent current in the first-order RC equivalent circuit of BESS;
[0170] Step 2.1.2: Construct a PV equivalent model;
[0171] From the perspective of power system modeling, the reactive power compensation demand characteristics of PV power generation systems under low voltage ride-through conditions are usually ignored, and the risk of relay protection malfunction caused by abnormal changes in grid current amplitude during transient processes is often ignored. The P-RLC model has a simple structure and strong generalization ability within a certain range. Therefore, the P-RLC model is selected as the PV equivalent sub-model for discretization. Furthermore, the power factor of the PV power generation system is close to 1, and its reactive power characteristics are not considered. The PV equivalent sub-model is shown in Equations (12)-(13).
[0172]
[0173] P PV =P0+I2|U R | (13)
[0174] In the formula: I2 is the current in the dynamic partial loop of the PV model; P PV U represents the overall output power of the model; R, L, and C represent the equivalent resistance, inductance, and capacitance values of the dynamic part of the PV model, respectively; U C For the dynamic portion of the capacitor voltage, U PV P0 is the grid terminal voltage, and P0 is the static active power output fixed value, such as... Figure 6 As shown;
[0175] Step 2.2: Discretization of the equivalent model;
[0176] Step 2.2.1: Construct the state equations and measurement equations of the sub-model;
[0177] When using the interactive multi-model algorithm for online modeling, the offline sub-model needs to be discretized; the state equations and measurement equations of the discretized sub-model are shown in equations (14)-(15);
[0178] x i (k)=F i (k-1)x i (k-1)+ω i (k-1) (14)
[0179] z i (k)=H i (k)x i (k)+v i (k) (15)
[0180] In the formula: x i (k), z i (k) represents the state vector and measurement vector of the i-th sub-model at time k; F i (k-1) is the state transition matrix of the i-th sub-model at time k-1; H i (k) is the measurement matrix of the i-th sub-model at time k; ω i (k-1) and v i (k) are the Gaussian white noise matrices of the i-th sub-model at times k-1 and k, respectively, with a mean of 0;
[0181] Step 2.2.2: Construct the discretization of IM;
[0182] The IM in the model adopts the third-order IM model, and its state equation is shown in equation (16);
[0183]
[0184] In the formula: E′ d E′ q The transient potentials on the d and q axes are respectively; X and X′ are the synchronous reactance and transient reactance, respectively; T′0 is the transient open-circuit time constant; w N ,w m These are the rated speed and rotor speed, respectively; s is the slip rate; i d i q These are the transient current components on the d and q axes, respectively; T e ,T m These are the electromagnetic torque and mechanical torque of the rotor, respectively; H is the inertial time constant.
[0185] The measurement equations for the third-order induction motor model are shown in equation (17);
[0186]
[0187] in,
[0188]
[0189] In the formula: P mQ m These represent the active and reactive power of IM, respectively; U is the voltage value of the load node; R s For IM stator resistance;
[0190] Step 2.2.3: Discretization of the BESS equivalent model;
[0191] Discretize equations (10) and (11) according to the forms of equations (14) and (15) to obtain the state equation of the BESS equivalent model as shown in equation (24) and the measurement equation as shown in equation (25). The specific derivation process is as follows:
[0192]
[0193] In the formula, T is the unit time interval;
[0194] The state equations of the BESS equivalent model are thus obtained, as shown in the following equation:
[0195]
[0196]
[0197] Since the values of each term in matrix Y approach zero, they are equivalent to noise terms. Finally, the simplified state equations and measurement equations of the BESS equivalent model are as follows:
[0198]
[0199] In the formula: P ES (k) represents the output power of BESS at time k, ω BESS v ES Both are Gaussian white noise in the state equations, U ES (k) represents the output voltage, and Z represents the parallel impedance (e.g., ...). Figure 6 (as shown);
[0200] Step 2.2.4: Discretization of the PV equivalent model;
[0201] By deriving equations (12) and (13), the dynamic state equation and measurement equation of the PV equivalent model are obtained as shown in equations (33) and (34). The specific derivation process is as follows:
[0202]
[0203] In the formula, U R U L These represent the resistor voltage and capacitor voltage in the generator dynamic model, respectively.
[0204] In steady state, the dynamic partial capacitor voltage U CEqual to the grid terminal voltage U PV The internal current is zero.
[0205]
[0206]
[0207] In the formula, U′ C (0) represents the transient port voltage; u C (t) is the capacitor port voltage at time t; I2(t) is as follows: Figure 6 , represents the dynamic partial loop current at time t; I2(0) represents the dynamic partial loop current at the initial time; I2(k+1) represents the dynamic partial loop current at time k+1 (corresponding to equations 14 and 15); f=50H Z ;s 1,2 Let s1 and s2 be as shown in Equation 31; U C (k) represents the terminal voltage at time k (corresponding to equations 14 and 15); K1 is a constant coefficient obtained from solving the differential equation; B is shown in equation 35; v PV ω PV Gaussian white noise; The initial voltage at the capacitor port; The current in the dynamic partial loop at the initial moment;
[0208] Furthermore, Step 3 includes:
[0209] As input for generalized load modeling, online power system data specifically includes the voltage amplitude U of load nodes. (k) The system frequency f, the total active power p injected into the node, the reactive power Q, the impedance value z, the state vectors of each sub-model, and Gaussian white noise, etc., need to be preprocessed before input interaction. This includes data cleaning to remove outliers and handle communication interruptions, as well as suppressing measurement noise through digital filtering technology, and finally forming a clean and aligned real-time measurement sequence Z(k), which serves as the direct input for subsequent model filtering and state updates.
[0210] Step 3.1, Input Interaction;
[0211] At time k-1, each sub-model corresponds to an optimal state vector x. i (k-1) and estimated covariance matrix P i (k-1), and then the state vectors and covariance matrices of each sub-model are weighted and fused according to the correctness probability obtained from the Markov operation matrix to obtain the fused optimal state vector x. o (k-1) and covariance matrix P o (k-1):
[0212]
[0213] In the formula, λ IM (k-1), λ ZIP (k-1), λ PV (k-1), λ BESS (k-1), λ i (k-1) is the state transition matrix of the i-th sub-model at time k-1; x IM (k-1), x ZIP (k-1), x PV (k-1), x BESS (k-1) represents the state vector of the i-th sub-model at time k-1; n is the number of generalized load types, including energy storage, photovoltaic, and generator.
[0214] Step 3.2, Model Filtering;
[0215] The fused x o (k-1) and P o (k-1) is used as the input variable, and an unscented Kalman filter is used for filtering and updating. The predicted states of each sub-model are then determined. and predicted covariance As shown in equations (38)-(39);
[0216]
[0217] in:
[0218]
[0219] In the formula, K i (k) represents the filter gain; s i (k) represents the residual covariance; Z(k) represents the fitted actual value; r i (k) represents the standard deviation of the measurement noise; Predict the state and prediction covariance for each sub-model; F represents the predicted state of the fused overall model. i Markov matrix coefficients; Q i A randomly generated n*n standard normal distribution random number matrix with mean 0 and variance 1; H i (k) is the measurement matrix of the i-th sub-model at time k;
[0220] Step 3.3: Probability Update;
[0221] The maximum likelihood function method is adopted to calculate the weight of each sub-model by the similarity between the current model and the target state. The maximum likelihood function is calculated as shown in Equation (41).
[0222]
[0223] The weight calculation formulas for each sub-model after the update are shown in equation (42);
[0224]
[0225] Step 3.4, Interactive Output;
[0226] The updated state is fused and estimated based on the latest weights of each sub-model to obtain the final output result;
[0227]
[0228] Furthermore, Step 4 includes:
[0229] Step 4.1, Dynamic tracking and detection of total system load power: Based on the weights u of each sub-model at time k obtained from the interactive output in Step 3. i (k) and its measurement output Z i (k), through weighted fusion, calculates and outputs the tracking and detection values of active power and reactive power of the generalized load of the system in real time;
[0230] Step 4.2, Online Identification and Detection of Generalized Load Composition Ratio: Directly record and output the weight probability set {u1(k), u2(k), i3(k)…u} of each sub-model at time k obtained through probability update in Step 3. n (k)}, and use this set as the detection result of the real-time composition ratio of various generalized loads in the system;
[0231] Step 4.3, Online identification and detection of key internal state variables: Extract and output the state of charge ξ of the battery energy storage system sub-model from the optimal state vector x0(k) obtained after the interactive output in Step 3. SOC (t) and the detection output state components of the induction motor sub-model, the state of charge of the energy storage sub-model and the expression of the detection output state components of the induction motor sub-model are as follows:
[0232]
[0233] The present invention also provides a novel online detection and state tracking system for generalized load components of a power system, the system comprising: a module for executing the aforementioned novel online detection and state tracking method for generalized load components of a power system.
[0234] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the novel online detection and state tracking method for generalized load components in a power system.
[0235] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the novel online detection and state tracking method for generalized load components in a power system.
[0236] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the novel online detection and state tracking method for generalized load components in a power system.
[0237] The present invention also includes Step 5: Case analysis and verification.
[0238] The specific steps of Step 5 are as follows:
[0239] Step 5.1: Based on the IEEE-33 node distribution network, some pure load nodes in the original model are replaced with different types of generalized load elements to verify the correctness of the proposed new power system generalized load online model and solution method based on interactive multi-model. At the same time, in order to verify the reliability of the simulation, two simulation scenarios are set in the example.
[0240] Scenario 1: Focusing on the issue of the single structure of the offline sub-model of the IMM algorithm, the offline sub-model of the IMM algorithm has been extended. It is necessary to verify the correctness of the established sub-model and its superiority or inferiority relationship with the existing model. Three different models are set up for comparative analysis.
[0241] Scenario 2: Since online model building methods have certain regional characteristics, in actual situations, the generalized load ratio or system structure may change due to various reasons. When comparing scenarios with changes in the generalized load ratio or system structure, the corresponding model verification is performed by setting up disconnection.
[0242] The IEEE-33 node distribution network model replaces some pure load nodes in the original model with different types of generalized load elements. The presence of each generalized load at a node indicates the type of generalized load present, not the same capacity. In the newly established power system simulation model, PV power accounts for 31%, wind power 14%, BESS capacity 10%, and traditional generator units 20%. Active loads are set at 10% of the total system capacity.
[0243] The main purpose of the simulation is to study the ability of the established time-varying online model to perform online real-time detection of generalized load components in a new power system, dynamic tracking of total load power, and online identification of key internal state variables. A short-circuit fault is set at point B0, causing the bus voltage to drop by 20%, and simulation operation data before and after the voltage drop are obtained. The fault data is measured at node 1, which is closest to the fault point. The discretized sub-model set is then used as the input of the sub-model to the IMM algorithm, and generalized load online modeling is performed according to equations (1)-(43).
[0244] Step 5.2, Scenario 1 Result Analysis;
[0245] Step 5.2.1, Analysis of power oscillation and fitting accuracy;
[0246] Table 1 shows the comparison of power output errors.
[0247]
[0248] Depend on Figure 3 From Table 1, we can draw the following conclusions:
[0249] 1) When using the traditional offline model, the fitting effect is relatively good at steady state. However, at voltage dips, due to the fixed structure of the offline model and the inability to use online real-time information for model correction, it is difficult to accurately reflect the fluctuation of system power at voltage dips. It cannot effectively fit the dynamic characteristics of the new power system after the generalized load is connected, resulting in the largest maximum error among the three scenario settings, and the overall fitting effect is also worse than the two online models.
[0250] 2) When using the two online models, the analysis of the discrete data time shows that when the new power system generalized load model with different sub-models is used to model the simulation system online, the maximum fitting error of active power and reactive power is smaller than that when the same sub-model is used. The maximum fitting error of active power is reduced by nearly 20%, which shows that the offline sub-model has a better fitting effect when the data fluctuates and can accurately reflect the dynamic characteristics of the new power system after the generalized load is connected.
[0251] 3) When using two online models, the analysis of the overall power output fitting shows that the root mean square error (RMSE) of active and reactive power is smaller when different sub-models are used, only around 1%, significantly less than the RMSE when using the same sub-model. The coefficient of determination for using different sub-models is also closer to 1 than that for using the same sub-model, indicating that the established offline sub-model performs better in overall power fitting. This high-precision power fitting result, while accurately detecting the total load components of the system online, can also record and output the real-time changes in the load composition ratio of each sub-model as a "generalized load composition detection report" for operators' reference.
[0252] Step 5.3, Scenario 2 Result Analysis;
[0253] Depend on Figure 4 The following conclusions can be drawn:
[0254] from Figure 4 As can be seen, the proposed novel generalized load model for power systems based on IMM has a good overall fit to changes in the proportion of generalized loads within the system and the system structure. When the power changes significantly, the proposed model, after an initial "inertial" oscillation process, re-tracks the new system power characteristics, verifying the ability of the proposed model to describe the dynamic changes in the load and structure of the novel power system. This indicates that the method described in this invention can still maintain stable and reliable detection capabilities when the system structure undergoes abrupt changes, and has strong robustness.
[0255] 1) Analysis from the perspective of online model tracking: Since the source-load ratio after node 5 is greater than that before node 5, after the line failure, the system output power is further reduced and the static load ratio increases, which suppresses the impact of oscillations caused by rotating elements and damping on the overall system, and causes the power fluctuation to decrease accordingly. This allows the IMM model to track the system characteristics more smoothly, so the fitting accuracy is relatively good.
[0256] 2) Analysis from the perspective of sub-models: Generalized loads within the same category share the same internal mathematical model structure and similar input-output characteristics. Using a capacity-weighted average method to aggregate generalized loads increases the weight of typical "users" with larger production scales within the overall "user" group, weakening the overall impact of smaller individuals. This results in sub-models with a certain degree of universal representativeness and strengthens their resilience to disturbances. Furthermore, the IMM model does not involve re-identifying the overall model parameters; instead, it uses the outputs of several sub-models to perform probability allocation and power fitting on the actual curves, further mitigating the impact of sub-model errors.
[0257] Step 5.4: Model applicability study;
[0258] Depend on Figure 5 and Figure 8 , Figure 9 From Table 2, we can draw the following conclusions:
[0259] To verify the reliability, accuracy, and applicability of the proposed method in actual online detection tasks, voltage drops of 10%, 30%, and 40% were set to study the model's adaptability. The power output under the different voltage drop conditions was fitted using the model constructed with a 20% voltage drop, and the power fitting error is shown in the figure below. Figure 5 and Figure 8 , Figure 9As shown in Table 2, the power output under different voltage drops was further compared and analyzed using the maximum error, root mean square error, and coefficient of determination.
[0260] Table 2 shows the comparison of power output errors.
[0261]
[0262] 1) As the voltage drop level increases, the fitting error values for active and reactive power also show an increasing trend. Furthermore, as shown in Table 1, the increase in active power error is greater than that for reactive power. This is because some components in the simulation system, such as PV and BESS, affect active power output, and the cumulative effect of active power error is stronger than that of reactive power error during voltage drop. The fitting error is smallest when the voltage drop is 10%, and the maximum power error is less than 0.06, indicating that the proposed model has good interpolation characteristics.
[0263] 2) From Figure 5 As can be seen, when the voltage drops by 30%, the power error does not increase significantly compared to when the voltage drops by 20%, and the coefficient of determination is greater than 95% for voltage drops within 30%, indicating that the proposed model has good externality for voltage drops within 30%. However, Figure 8 When the voltage drops by 40%, the active power error surges, with the maximum error even reaching 1.3 times the reference value, and from... Figure 8 The active power error can be seen from the data, which is much larger than... Figure 5 The error value of reactive power in the model can be attributed to the following reasons: 1) the inherent "inertia" effect of IMM; 2) when the voltage drop is too large, the PV equivalent model exhibits strong nonlinear characteristics, affecting the adaptability of its simplified model structure and the stability of its parameters, making the PV equivalent model no longer applicable. Furthermore, since the PV power factor is close to 1, it has a significant impact on active power but a relatively small impact on reactive power; 3) the sub-models in the optimal sub-model set are constructed offline using the statistical synthesis method. Their fixed structure and parameters limit the description of their respective generalized load dynamic characteristics, but their fitting accuracy is still higher than that of traditional offline modeling methods. Although the IMM algorithm can mitigate reasons 2) and 3) to some extent when assigning and updating the probability of each sub-model due to the weight relationship, the deviation increases with the increase of voltage drop, leading to an aggravation of the final online model fitting error.
[0264] In summary, the online detection and status tracking method proposed in this invention exhibits excellent detection accuracy and stability within a 30% voltage drop range, with an error range of less than 5%. It can provide highly reliable real-time status information for the safe operation of the power grid and possesses significant engineering application value. Furthermore, this study clarifies the boundaries of its detection capabilities, providing clear reliability guidance for engineering applications.
[0265] This invention relates to a novel online detection and state tracking method for generalized load components in power systems, belonging to the field of power system operation monitoring technology. It aims to solve the technical challenge of existing technologies being unable to perceive the dynamic composition and internal state of generalized loads in new power systems in real time. The invention includes the following steps: selecting generalized load units based on an optimized statistical clustering method; establishing and expanding an offline sub-model set including distributed generation and battery energy storage systems (BESS), and discretizing the expanded offline sub-model set; performing online modeling of generalized load components using online system data based on the interactive multi-model (IMM) algorithm; and using the established online model to achieve real-time online detection of generalized loads in new power systems, dynamic tracking of total load power, and online identification of key internal state variables. This invention not only boasts high model fitting accuracy and fast iteration speed but also enables real-time and accurate identification and early warning of generalized load anomalies, faults, or operational state deviations, providing crucial data support for the safe and stable operation of the system.
[0266] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.
Claims
1. A novel online detection and state tracking method for generalized load components in a power system, characterized in that: The method includes: Step 1: Select generalized load cells based on optimized statistical clustering method; Step 2: Establish and expand the offline sub-model set that includes distributed power sources and battery energy storage systems, and discretize the expanded offline sub-model set. Step 3: Based on the interactive multi-model algorithm, perform online modeling of generalized load using online power system data.
2. The novel online detection and state tracking method for generalized load components in a power system according to claim 1, characterized in that: The method further includes: Step 4: Based on the established generalized load online model, perform real-time detection of the generalized load composition of the new power system, dynamic tracking of total load power, and online identification of key internal state variables.
3. The novel online detection and state tracking method for generalized load components in a power system according to claim 1, characterized in that: Step 1 includes: Step 1.1: Construct the characteristic parameter matrix; Taking m equipment components in the power system as research objects, each object is represented by n characteristic data, thus establishing the characteristic parameter matrix U*, as shown in equation (1); Where, x mn This represents the nth feature data of the mth device component; The characteristic parameter matrix U* is formatted as shown in equation (2); In the formula, x′ mn x represents the formatted values of each element in U*. n This represents the nth feature data. The data is normalized to between 0 and 1 to eliminate the influence of dimensions; the fuzzy similarity relationship of the equipment components is determined by multivariate analysis, a similarity matrix R is established, and the transitive closure t(R1) of the similarity matrix is obtained by squaring, so that it satisfies the relationship of equation (3); R1→R1 2 →R1 4 →... (3) After performing a finite number of K / 2 operations on the above model, the transitive closure can be determined, as shown in equation (4); R1 K =t(R1)=R1 K ·R1 K (4) Step 1.2, Construct the domain of discourse for the basic unit device: Based on the analysis and research of the basic unit operating conditions during normal system operation, according to industry classification, several typical "users" that can reflect the production characteristics of each industry are selected for preliminary investigation based on practical experience; on the basis of the preliminary selection results, the generalized load units of "users" that are not typical in reality are removed by using the mathematical method of fuzzy clustering, and the generalized load characteristics are further integrated; the percentage of equipment parameters of each basic unit is calculated based on the statistical physical characteristic parameters of the equipment components, and the universe of discourse of the basic unit equipment is established, as shown in Equation (5): The same maximum value normalization method as Equation (2) is used to format each column of the universe of discourse L1; Step 1.3: Construct a fuzzy similarity matrix; The similarity coefficient r is introduced using the minimax method. mn , representing the m-th and n-th feature data of two samples, x m With x n To determine the degree of similarity between them, a fuzzy similarity matrix R1 is established, as shown in equation (6): The maximum-minimum method is shown in equation (7): In the formula, x nj ,x mj represent the nth column element and the mth row element in U*, respectively; ∧ and ∨ represent the intersection and union operations, respectively. Step 1.4: Construct an equivalent matrix that reflects the physical characteristics of each type of basic unit equipment; The transitive closure t(R1) of the similarity matrix is solved using equations (3)-(4), and the equivalent matrix reflecting the physical characteristics of each type of basic unit equipment is obtained, as shown in equation (8): By selecting an appropriate value of λ to construct the cutoff matrix, the fuzzy clustering results are obtained. When λ = 0.4, the cutoff matrix t(0.4) is shown in equation (9):
4. The novel online detection and state tracking method for generalized load components in a power system according to claim 1, characterized in that: Step 2 includes: Step 2.1: Construct a generalized load mathematical model; Step 2.1.1: Construct an equivalent BESS model; The BESS first-order RC equivalent circuit model is adopted, and the BESS first-order RC equivalent circuit model is shown in Equation (10). In the formula: U BESS (t), U oc (t) represent the port voltage and open-circuit voltage of the battery model at time t, respectively; R p U p and C p These are the battery's polarization internal impedance, voltage, and capacitance, respectively; R0 is the battery's equivalent internal resistance. The state-of-charge expression of BESS obtained by the ampere-hour integration method is shown in the figure below; In the formula: ξ SOC (t) represents the state of charge of BESS at time t; Q N η is the rated capacity of the battery; η is the coulomb efficiency coefficient; I1 is the equivalent current in the first-order RC equivalent circuit of BESS; Step 2.1.2: Construct a PV equivalent model; The P-RLC model is selected as the PV equivalent sub-model for discretization, and the power factor of the PV power generation system is close to 1, and its reactive power characteristics are not considered; the PV equivalent sub-model is shown in Equation (12)-Equation (13). P PV =P0+I2|U R | (13) In the formula: I2 is the current in the dynamic partial loop of the PV model; P PV U represents the overall output power of the model; R, L, and C represent the equivalent resistance, inductance, and capacitance values of the dynamic part of the PV model, respectively; U C U is the voltage across the dynamic capacitor. PV P0 is the grid terminal voltage, and P0 is the static active power output fixed value. Step 2.2: Discretization of the equivalent model; Step 2.2.1: Construct the state equations and measurement equations of the sub-model; When using the interactive multi-model algorithm for online modeling, the offline sub-model needs to be discretized; the state equations and measurement equations of the discretized sub-model are shown in equations (14)-(15); x i (k)=F i (k-1)x i (k-1)+ω i (k-1) (14) z i (k)=H i (k)x i (k)+v i (k) (15) In the formula: x i (k), z i (k) represents the state vector and measurement vector of the i-th sub-model at time k; F i (k-1) is the state transition matrix of the i-th sub-model at time k-1; H i (k) is the measurement matrix of the i-th sub-model at time k; ω i (k-1) and v i (k) are the Gaussian white noise matrices of the i-th sub-model at times k-1 and k, respectively, with a mean of 0; Step 2.2.2: Construct the discretization of IM; The IM in the model adopts the third-order IM model, and its state equation is shown in equation (16); In the formula: E′ d E′ q The transient potentials on the d and q axes are respectively; X and X′ are the synchronous reactance and transient reactance, respectively; T′0 is the transient open-circuit time constant; w N ,w m These are the rated speed and rotor speed, respectively; s is the slip rate; i d i q These are the transient current components on the d and q axes, respectively; T e ,T m These are the electromagnetic torque and mechanical torque of the rotor, respectively; H is the inertial time constant. The measurement equations for the third-order induction motor model are shown in equation (17); in, In the formula: P m Q m These represent the active and reactive power of IM, respectively; U is the voltage value of the load node; R s For IM stator resistance; Step 2.2.3: Discretization of the BESS equivalent model; Discretize equations (10) and (11) according to the forms of equations (14) and (15) to obtain the state equation of the BESS equivalent model as shown in equation (24) and the measurement equation as shown in equation (25). The specific derivation process is as follows: In the formula, T is the unit time interval; The state equations of the BESS equivalent model are thus obtained, as shown in the following equation: Since the values of each term in matrix Y approach zero, they are equivalent to noise terms. Finally, the simplified state equations and measurement equations of the BESS equivalent model are as follows: In the formula: P ES (k) represents the output power of BESS at time k, ω BESS v ES Both are Gaussian white noise in the state equations, U ES (k) represents the output voltage, and Z represents the parallel impedance; Step 2.2.4: Discretization of the PV equivalent model; By deriving equations (12) and (13), the dynamic state equation and measurement equation of the PV equivalent model are obtained as shown in equations (33) and (34). The specific derivation process is as follows: In the formula, U R U L These represent the resistor voltage and capacitor voltage in the generator dynamic model, respectively. In steady state, the dynamic partial capacitor voltage U C Equal to the grid terminal voltage U PV The internal current is zero. In the formula, U′ C (0) represents the transient port voltage; u C (t) represents the capacitor port voltage at time t; I2(t) represents the dynamic partial loop current at time t; I2(0) represents the dynamic partial loop current at the initial time; I2(k+1) represents the dynamic partial loop current at time k+1; f = 50H Z ;s 1,2 Let s1 and s2 be as shown in Equation 31; U C (k) represents the terminal voltage at time k; K1 is a constant coefficient obtained from solving the differential equation; v PV ω PV All are Gaussian white noise; The initial voltage at the capacitor port; The current in the dynamic partial loop is at the initial moment.
5. A novel online detection and state tracking method for generalized load components in a power system according to claim 1, characterized in that: Step 3 includes: Step 3.1, Input Interaction; As input for generalized load modeling, online power system data specifically includes the voltage amplitude U of load nodes. (k) The system frequency f, the total active power p injected into the node, the reactive power Q, the impedance value z, the state vector of each sub-model, and Gaussian white noise are preprocessed before input interaction. This includes data cleaning to remove outliers and handle communication interruptions, as well as suppressing measurement noise through digital filtering technology. Finally, a clean and aligned real-time measurement sequence Z(k) is formed, which serves as the direct input for subsequent model filtering and state updates. At time k-1, each sub-model corresponds to an optimal state vector x. i (k-1) and estimated covariance matrix P i (k-1), and then the state vectors and covariance matrices of each sub-model are weighted and fused according to the correctness probability obtained from the Markov operation matrix to obtain the fused optimal state vector x. o (k-1) and covariance matrix P o (k-1): In the formula, λ IM (k-1), λ ZIP (k-1), λ PV (k-1), λ BESS (k-1), λ i (k-1) is the state transition matrix of the i-th sub-model at time k-1; x IM (k-1), x ZIP (k-1), x PV (k-1), x BESS (k-1) represents the state vector of the i-th sub-model at time k-1; n is the number of generalized load types, including energy storage, photovoltaic, and generator. Step 3.2, Model Filtering; The fused x o (k-1) and P o (k-1) is used as the input variable, and an unscented Kalman filter is used for filtering and updating. The predicted states of each sub-model are then determined. and predicted covariance As shown in equations (38)-(39); in: In the formula, K i (k) represents the filter gain; s i (k) represents the residual covariance; Z(k) represents the fitted actual value; r i (k) represents the standard deviation of the measurement noise; Predict the state and prediction covariance for each sub-model; F represents the predicted state of the fused overall model. i Markov matrix coefficients; Q i A randomly generated n*n standard normal distribution random number matrix with mean 0 and variance 1; H i (k) is the measurement matrix of the i-th sub-model at time k; Step 3.3: Probability Update; The maximum likelihood function method is adopted to calculate the weight of each sub-model by the similarity between the current model and the target state. The maximum likelihood function is calculated as shown in Equation (41). The weight calculation formulas for each sub-model after the update are shown in equation (42); Step 3.4, Interactive Output; The updated state is fused and estimated based on the latest weights of each sub-model to obtain the final output result; 6. A novel online detection and state tracking method for generalized load components in a power system according to claim 2, characterized in that: Step 4 includes: Step 4.1, Dynamic tracking and detection of total system load power: Based on the weights u of each sub-model at time k obtained from the interactive output in Step 3. i (k) and its measurement output Z i (k), through weighted fusion, calculates and outputs the tracking and detection values of active power and reactive power of the generalized load of the system in real time; Step 4.2, Online Identification and Detection of Generalized Load Composition Ratio: Directly record and output the weight probability set {u1(k), u2(k), u3(k) … u} of each sub-model at time k obtained through probability update in Step 3. n (k)}, and use this set as the detection result of the real-time composition ratio of various generalized loads in the system; Step 4.3, Online identification and detection of key internal state variables: Extract and output the state of charge ξ of the battery energy storage system sub-model from the optimal state vector x0(k) obtained after the interactive output in Step 3. SOC (t) and the detection output state components of the induction motor sub-model, the state of charge of the energy storage sub-model and the expression of the detection output state components of the induction motor sub-model are as follows:
7. A novel online detection and condition tracking system for generalized load components in a power system, characterized in that, The system includes a module for performing a novel online detection and state tracking method for generalized load components in a power system as described in any one of claims 1 to 6.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a novel online detection and state tracking method for generalized load components in a power system as described in any one of claims 1 to 6.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a novel online detection and state tracking method for generalized load components in a power system as described in any one of claims 1 to 6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements a novel online detection and state tracking method for generalized load components in a power system as described in any one of claims 1 to 6.
