Inertia evaluation method and device for electric hydrogen energy hub

Abstract

This invention relates to a method and apparatus for assessing the inertia of an electric-hydrogen energy hub, specifically in the field of assessing the inertia response capability of an electric-hydrogen fusion hub. The method includes: Step S1: Determining the inertia center, using the data measured by the synchronous phasor unit at that point as the data basis, and preprocessing the monitored data such as standardization, filtering, and detrending; Step S2: Establishing a physical model based on the inertia characteristics of the synchronous generator and the virtual inertia control method to estimate the synchronous and asynchronous inertia respectively; Step S3: Establishing a data model for an extreme learning machine, determining its framework, and then using an ant colony algorithm to train the data model to obtain optimal parameters. The physical model and the data model are then combined in a serial manner to form a data-physical fusion model. This invention addresses the issue of low accuracy in physical models by incorporating an extreme learning machine data model, proposing an inertia estimation scheme based on data-physical fusion.

Description

Technical Field

[0001] This invention relates to the field of evaluating the inertia response capability of an electric-hydrogen fusion hub, specifically to a method and apparatus for evaluating the inertia of an electric-hydrogen energy hub. Background Technology

[0002] The deployment of high-penetration renewable energy and the integration of hydrogen energy technologies will face the significant challenge of continuously decreasing inertia. The inertia of traditional power systems is primarily related to the rotating mass of synchronous generators, which helps maintain system frequency stability by releasing stored energy to resist frequency changes. In power-hydrogen integration hubs, most units are connected to the grid via power electronic converters, which do not provide fixed inertia support. Therefore, with the increasing share of new energy and hydrogen power generation, the overall system inertia decreases, making the system more susceptible to frequency stability issues. This decrease in system inertia leads to a high rate of frequency change (RoCoF), resulting in sharp frequency drops or rises and larger frequency deviations. This situation triggers RoCoF and low-frequency load shedding relays, and in the worst case, can cause cascading tripping of generators, leading to a system-wide blackout.

[0003] While inertia fluctuations are not significant in traditional systems with a large proportion of synchronous generators, they have become an increasingly concerning issue in hybrid electric-hydrogen systems where inertia is gradually decreasing. Therefore, accurately estimating the available network inertia is crucial for system operators and grid planning. Accurate inertia estimates enable appropriate countermeasures to overcome frequency stability challenges. These measures include incorporating appropriate virtual inertia control and deploying suitable energy storage devices based on the system's inertia level.

[0004] It is worth noting that, to address the challenges of low-inertia systems, inverter interfaces and battery storage systems may be configured with corresponding virtual inertia support. Traditional inertia estimation can only capture the synchronous inertia of grid-connected generation and loads. With the introduction of virtual inertia, the accuracy of traditional estimation methods decreases. Therefore, new inertia estimation methods are needed to more accurately assess the overall system inertia. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method and apparatus for inertia assessment of an electric-hydrogen energy hub based on data-physical fusion. This invention offers an inertia assessment method in the context of increasing new energy penetration and the integration of power systems with hydrogen energy. Data obtained from synchronous phasor measurement units (PMUs) is used as the data foundation, and the inertia center is determined for data preprocessing. A physical model for estimating the synchronous generator inertia is determined based on the synchronous generator's state. Simultaneously, a physical model for calculating the overall inertia of the system is obtained by combining this model with a physical model established using virtual inertia support methods.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for assessing the inertia of an electric hydrogen energy hub includes the following steps:

[0008] The center of inertia is determined, and the data measured by the synchronous phasor unit of the center of inertia is used as the data basis. The measured data is then normalized, filtered, and detrended preprocessed.

[0009] Based on the characteristics of synchronous generator inertia and the virtual inertia control method, a physical model of synchronous generator inertia is established. The preprocessed data is substituted into the physical model to estimate the synchronous inertia and asynchronous inertia respectively, and the total inertia of the preliminary hydrogen energy hub is obtained by superposition.

[0010] A data model for inertia assessment is established for the Extreme Learning Machine (ELM). After determining the framework of the ELM, the data model is trained using the ant colony algorithm to obtain the optimal parameters. The physical model and data model of the synchronous generator inertia are combined in a serial manner to form a data-physical fusion model. The total inertia obtained from the physical model is input into the data model for correction to obtain the total inertia of the hydrogen-electric energy hub.

[0011] Further, step S1 specifically includes:

[0012] By introducing the center of inertia to select the optimal frequency sampling point, and using formula (1) to traverse each node in the system, the optimal frequency sampling point is found. E The node with the smallest value is used as the frequency sampling point.

[0013]

[0014] Among them, F E The cost function value; u represents the node number; i represents the generator number; H i S i Let be the inertial constant and rated capacity of generator i, respectively; d 2 (u,i) represents the squared distance between node u and generator node i; u∈m represents the u-th node in the system node set;

[0015] Based on the data obtained by the synchronous phasor measurement device at the optimal frequency sampling point, the data is normalized and mapped to the [0,1] interval. The normalization method is shown in formula (2):

[0016]

[0017] Where, x i_max and x i_min Let x represent the maximum and minimum values ​​of the i-th data set. i *This represents the data after normalization; x i This represents data that has not undergone normalization.

[0018] The components of the system inertial response are obtained by isolating the normalized data using a Butterworth low-pass filter.

[0019] Perform DFA detrending.

[0020] Furthermore, the DFA detrending includes:

[0021] The original signal is integrated to remove the DC offset present in the data;

[0022] The integrated signal is divided into multiple windows or boxes of variable length n, and then a least-squares first-order linear approximation is performed on each window to represent the "trend" of the signal segment.

[0023] Calculate the root mean square (RMS) of the detrending signal, determine the location of the event based on the RMS, set an appropriate RMS based on the network characteristics, determine the start time of the event, assess the applicability of the event for inertia estimation, and complete the DFA detrending.

[0024] Further, step S2 specifically includes:

[0025] Step S21: Analyze the operating characteristics of the synchronous generator and determine the physical model for estimating the synchronous generator's inertia based on the generator's state.

[0026]

[0027] Among them, H sync H is the equivalent inertia constant of the synchronous generator. Gk Let x be the inertia constant of the k-th synchronous generator. Gk This represents the start / stop status of generator k, where n is the number of synchronous generators;

[0028] Step S22: Establish a corresponding physical model based on the control methods provided for virtual inertia, for virtual inertia estimation;

[0029] For the low-order system model based on droop control, as shown in Equation (13):

[0030]

[0031] Among them, H droop (s) is the transfer function of the low-order system model with droop control, where T1 and T d1 H represents the time delay and time constant corresponding to droop control. droop Gain for droop control;

[0032] For the low-order system model based on RoCoF control, as shown in equation (14):

[0033]

[0034] Among them, H RoCoF (s) is the transfer function of the low-order system model based on RoCoF control, T2 and T d2 To determine the time delay and time constant corresponding to RoCoF control, H RoCoF Gain based on RoCoF control;

[0035] With the help of system identification, virtual inertia contribution TF Non-Sync The overall equivalent transfer function is:

[0036]

[0037] Among them, △P Non-Sync The power provided for the virtual inertia, Δf is the system frequency imbalance;

[0038] The overall virtual inertia transfer function is:

[0039]

[0040] Among them, H Non-Sync (s) is the transfer function of the system's virtual inertia, T d For equivalent time delay;

[0041] Step S23: Obtain the virtual inertia by identifying system parameters based on the overall virtual inertia transfer function, and combine it with the synchronous generator inertia to obtain the system inertia:

[0042] Substitute the PMU data processed in step S12 into the total power imbalance and frequency imbalance of the system. Determine the imbalance borne by the synchronous generator by the magnitude of the synchronous generator's inertia, as shown in formula (17):

[0043]

[0044] Among them, △P Sync The imbalance borne by the synchronous generator; The rate of change of frequency imbalance;

[0045] The power provided by the virtual inertia source is separated from the total power imbalance, as shown in Equation (18):

[0046] ΔP Non-Sync =ΔP - ΔP Sync (18)

[0047] Where ΔP is the system power imbalance;

[0048] The virtual equivalent inertia is obtained based on the power provided by the separated virtual inertia source, the system frequency imbalance, and the virtual equivalent inertia transfer function.

[0049] The system inertia is equal to the superposition of the synchronous generator inertia and the virtual inertia, as shown in formula (19):

[0050] H sys =H Non-Sync +H Sync (19)

[0051] Among them, H sys H represents the overall inertia of the system. Non-Sync This is the virtual equivalent inertia.

[0052] Furthermore, step S3 specifically includes:

[0053] Step S31: Select the input features to determine the number of neurons in the input layer of the extreme learning machine;

[0054] Step S32: Establish the basic structural framework of the Extreme Learning Machine and use the ant colony algorithm to optimize the weights and thresholds of the input layer and hidden layer of the Extreme Learning Machine;

[0055] Step S33: Connect the physical model and the data model sequentially to obtain the data-physical fusion inertia estimation model, establish an evaluation standard model, and determine the accuracy of the data-physical fusion inertia estimation model:

[0056] The accuracy of the estimation results is judged using the following evaluation model, with the following evaluation indicators:

[0057] Root Mean Square Error (RMSE):

[0058]

[0059] Among them, H r,l The actual value of inertia in the test data; H p,l The inertia prediction value for the test data;

[0060] When the root mean square error meets expectations, the data-physical fusion inertia estimation model can be applied to real-time inertia estimation.

[0061] Further, step S32 includes:

[0062] For a specific training sample set of N sets of data (x p ,t p ), x p =[x p1 ,x p2 ,…,x pn ]T ∈R n The input vector includes the system's equivalent inertia estimation results from the physical model, the location of the perturbation point, the location of the sampling point, frequency data, and power data; t p =[t p1 ,t p2 ,…,t pm ] T ∈R m The output vector includes the corrected equivalent inertia calculation results;

[0063] The mathematical model of the Extreme Learning Machine is expressed as:

[0064]

[0065] Where M is the number of hidden layer units; g(x) is the activation function; w q =[w q1 ,w q2 ,…,w qn ] T β is the weight vector connecting the input vector and the q-th hidden layer unit; q =[β q1 ,β q2 ,…,β qn ] T b is the weight vector connecting the q-th hidden layer unit and the output vector; q The bias is set for the q-th hidden layer unit; o j =[o j1 ,o j2 ,…,o jm ] T is the predicted output vector; j is the j-th data set; N is the total number of data sets; the superscript T indicates the transpose of the matrix;

[0066] If the prediction result is o j It can accurately approximate the actual results t of these N sets of samples j Then the following equation holds:

[0067]

[0068] Among them, o j For the prediction result, t j For the actual result; ||o j -t j || represents the norm of the difference in the results;

[0069] Substitute formula (21) into formula (20):

[0070]

[0071] Equation (22) can be represented as a matrix:

[0072] Hβ=T (23)

[0073] in:

[0074]

[0075]

[0076]

[0077] Where H is the hidden layer output matrix of the neural network; β is the output weight matrix, which is an important parameter for establishing the error fitting model; and T is the prediction result of the frequency situation characteristics.

[0078] In the Extreme Learning Machine algorithm, w is determined randomly. i and b i Its hidden layer output matrix H is then uniquely determined; according to the linear system problem Hβ=T, the output weight β can be approximately obtained by the following formula:

[0079]

[0080] in, Let H be the generalized inverse matrix of H.

[0081] Further, step S32 includes: after establishing the basic framework of the Extreme Learning Machine, optimizing and training it based on the ant colony algorithm, including:

[0082] Step 1. Determine the topology of the Extreme Learning Machine, namely the number of neurons in the input layer, the number of neurons in the hidden layer, and the number of neurons in the output layer;

[0083] Step 2. Encode the weights and thresholds from the input layer to the hidden layer in the extreme learning machine to obtain the initial population;

[0084] Step 3. Decode the weights and thresholds, and feed them into the training network of the Extreme Learning Machine. Use the training samples to train the machine.

[0085] Step 4. After training is complete, test the sample and use the sum of squared errors between the expected value and the predicted value of the test sample as the fitness function.

[0086] Step 5. Select, crossover, and mutate the population to obtain a new population. If the conditions are met, the network weights and thresholds with the minimum sum of squared errors are obtained. If the conditions are not met, return to step 2.

[0087] Step 6. Input the optimized weights and thresholds into the training network, calculate the hidden layer output matrix H, and solve for the generalized inverse matrix of H.

[0088] Step 7. Calculate the output layer weights using formula (27);

[0089] Step 8. Feed the test samples into the Extreme Learning Machine model for prediction.

[0090] The present invention also provides an inertia assessment device for an electric hydrogen energy hub, comprising:

[0091] The data processing module is used to read the raw data measured by the synchronous phasor measurement device, and to normalize the raw data to make it dimensionless for easy calculation. Then, filtering technology is used to eliminate problems such as oscillation and noise in the raw data, and the events that can be used for inertia estimation are determined by detrending. The obtained data serves as the data basis for the subsequent modules to reduce estimation errors.

[0092] The physical model inertia estimation module, based on the synchronous generator inertia characteristics that are only related to the start-up state and the virtual inertia characteristics that are controlled based on RoCoF, frequency deviation, or a combination of both, finds the relationship between the input quantity and inertia, constructs a flowchart, and calculates the transfer function. Based on the preprocessed data, it performs preliminary synchronous inertia and virtual inertia estimation, and superimposes them to obtain the total inertia.

[0093] The data model inertia estimation module determines the number of neurons in the input layer of the Extreme Learning Machine (ELM) and uses the ant colony algorithm to optimize the weights and thresholds of the input and hidden layers of the ELM to obtain the ELM model. The inertia initially estimated by the physical model is substituted into the data model for correction to obtain the overall inertia of the hydrogen-electric energy hub.

[0094] Furthermore, in the data processing module, the optimal frequency sampling point is selected by introducing the inertia center, and the optimal frequency sampling point is found by traversing each node in the system using formula (1). E The node with the smallest value is used as the frequency sampling point.

[0095]

[0096] Among them, F E The cost function value; u represents the node number; i represents the generator number; H i S i Let be the inertial constant and rated capacity of generator i, respectively; d 2 (u,i) represents the squared distance between node u and generator node i; u∈m represents the u-th node in the system node set;

[0097] Based on the data obtained by the synchronous phasor measurement device at the optimal frequency sampling point, the data is normalized and mapped to the [0,1] interval. The normalization method is shown in formula (2):

[0098]

[0099] Where, x i_max and x i_min Let x represent the maximum and minimum values ​​of the i-th data set. i * This represents the data after normalization; x i This represents data that has not undergone normalization.

[0100] The components of the system inertial response are obtained by isolating the normalized data using a Butterworth low-pass filter.

[0101] Perform DFA detrending.

[0102] Furthermore, the physical model inertia estimation module implements:

[0103] Analyze the operating characteristics of the synchronous generator and determine the physical model for estimating the generator's inertia based on its state:

[0104]

[0105] Among them, H sync H is the equivalent inertia constant of the synchronous generator. Gk Let x be the inertia constant of the k-th synchronous generator. Gk This represents the start / stop status of generator k, where n is the number of synchronous generators;

[0106] A corresponding physical model is established based on the control methods provided for virtual inertia, which is used for virtual inertia estimation;

[0107] For the low-order system model based on droop control, as shown in Equation (13):

[0108]

[0109] Among them, H droop (s) is the transfer function of the low-order system model with droop control, where T1 and T d1 H represents the time delay and time constant corresponding to droop control. droop Gain for droop control;

[0110] For the low-order system model based on RoCoF control, as shown in equation (14):

[0111]

[0112] Among them, H RoCoF (s) is the transfer function of the low-order system model based on RoCoF control, T2 and T d2 To determine the time delay and time constant corresponding to RoCoF control, H RoCoF Gain based on RoCoF control;

[0113] With the help of system identification, virtual inertia contribution TF Non-Sync The overall equivalent transfer function is:

[0114]

[0115] Among them, △P Non-Sync The power provided for the virtual inertia, Δf is the system frequency imbalance;

[0116] The overall virtual inertia transfer function is:

[0117]

[0118] Among them, H Non-Sync (s) is the transfer function of the system's virtual inertia, T d For equivalent time delay;

[0119] The virtual inertia is obtained by identifying system parameters based on the overall virtual inertia transfer function, and then combined with the synchronous generator inertia to obtain the system inertia:

[0120] Substitute the processed data from the synchronous phasor measurement device to calculate the total power imbalance and frequency imbalance of the system. Determine the imbalance borne by the synchronous generator based on the magnitude of the synchronous generator's inertia, as shown in formula (17):

[0121]

[0122] Among them, △P Sync The imbalance borne by the synchronous generator; The rate of change of frequency imbalance;

[0123] The power provided by the virtual inertia source is separated from the total power imbalance, as shown in Equation (18):

[0124] ΔP Non-Sync =ΔP - ΔP Sync (18)

[0125] Where ΔP is the system power imbalance;

[0126] The virtual equivalent inertia is obtained based on the power provided by the separated virtual inertia source, the system frequency imbalance, and the virtual equivalent inertia transfer function.

[0127] The system inertia is equal to the superposition of the synchronous generator inertia and the virtual inertia, as shown in formula (19):

[0128] H sys =H Non-Sync +H Sync (19)

[0129] Among them, H sys H represents the overall inertia of the system. Non-Sync This is the virtual equivalent inertia.

[0130] Compared with the prior art, the present invention has the following advantages:

[0131] 1. For applications in different scenarios, this invention can accurately capture the specific changes in the transient state of system inertia support, ensuring the reliability of input data.

[0132] 2. The virtual inertia response has a certain time delay, which is different from the response characteristics of the synchronous generator. The filtering stage in the inverter circuit also causes time delay. At the same time, changes in the inverter topology will also affect the virtual inertia response. Simply substituting the power imbalance ΔP and the rate of change of frequency RoCoF at the beginning of the disturbance into the oscillation equation, as in traditional estimation methods, will not accurately capture the system inertia constant. Considering the characteristics of virtual inertia, this invention estimates the synchronous generator inertia and virtual inertia separately. The virtual inertia estimation is based on the physical model established by the virtual inertia control method, which can well take into account the system time delay and topology changes, thus improving the estimation accuracy.

[0133] 3. While physical methods are simple to apply and computationally fast, they neglect some known and unknown factors during modeling, leading to systematic errors and low computational accuracy. Data models, on the other hand, can handle complex problems, but they are highly dependent on data scale and quality, and the results lack interpretability. Therefore, this invention constructs a data-physical fusion model, where the physical model can constrain the data, making the model interpretable, while the data model can improve the overall accuracy of the system. Using a data-physical fusion model can further improve the accuracy of inertia estimation.

[0134] 4. Extreme Learning Machines (ELMs) do not require iterative adjustments to their hidden layers during the learning process, exhibit fast learning speed, and strong generalization ability. Furthermore, their weight parameter calculation method guarantees globally optimal results. Combining this with the ant colony algorithm avoids the overfitting phenomenon that occurs during training due to excessive hidden layer nodes in traditional ELMs. The weights and thresholds of the input and hidden layers are optimized, thereby improving the model's stability and accuracy. Attached Figure Description

[0135] Figure 1 This is a flowchart of a method for evaluating the inertia of an electric hydrogen energy hub according to an embodiment of the present invention;

[0136] Figure 2 This is a block diagram of the droop control in an embodiment of the present invention;

[0137] Figure 3 This is a RoCoF-based control block diagram in an embodiment of the present invention;

[0138] Figure 4 This is a block diagram of the droop-RoCoF fusion control in an embodiment of the present invention;

[0139] Figure 5 This is a schematic diagram of the Extreme Learning Machine structure in an embodiment of the present invention.

[0140] Figure 6 This is a schematic diagram of a virtual device in an embodiment of the present invention. Detailed Implementation

[0141] This invention provides a data-physical fusion-based inertia estimation method for electric hydrogen energy hubs. In the context of increasing new energy penetration and the integration of power systems with hydrogen energy, this invention aims to alleviate the problem of low accuracy of single physical models by fusing the system's physical model with the data model of an extreme learning machine, thus proposing an inertia estimation method based on data-physical fusion.

[0142] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0143] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for evaluating the inertia of an electric hydrogen energy hub, comprising the following steps:

[0144] Step S1: Determine the center of inertia, and use the data measured by the synchronous phasor unit of the center of inertia as the data basis. Perform preprocessing such as per-unitization, filtering and detrending on the measured data.

[0145] Step S2: Establish a physical model of the synchronous generator's inertia based on the characteristics of the synchronous generator's inertia and the virtual inertia control method, and estimate the synchronous inertia and asynchronous inertia respectively;

[0146] Step S3: Establish a data model for the Extreme Learning Machine. After determining the framework of the Extreme Learning Machine, use the ant colony algorithm to train the data model to obtain the optimal parameters, and combine the physical model and the data model in a serial manner to form a data-physical fusion model.

[0147] In this embodiment, step S1 specifically includes:

[0148] Step S11: After the disturbance, the system frequency has spatial distribution characteristics. The different frequency sampling points will directly affect the final calculation results. Therefore, the inertia center is introduced to select the optimal frequency sampling point.

[0149] Since the inertia in the system is distributed, it is necessary to determine the frequency sampling points. This can be done by iterating through each node in the system using formula (1) to find the Fs. E The node with the smallest value is used as the frequency sampling point.

[0150]

[0151] Among them, F E The cost function value; u represents the node number; i represents the generator number; H i S i Let be the inertial constant and rated capacity of generator i, respectively; d 2 (u,i) represents the squared distance between node u and generator node i; u∈m represents the u-th node in the system node set.

[0152] Step S12: Based on the data obtained by the PMU at the optimal frequency sampling point, different types of data have different units, so the data needs to be normalized. At the same time, filtering is used to eliminate problems such as oscillation and noise in the original data, and events that can be used for inertia estimation are determined by detrending.

[0153] To avoid calculation errors caused by large differences in the order of magnitude between different categories of data, it is necessary to normalize the measured data and map it to the interval [0,1]. The normalization method is shown in formula (2).

[0154]

[0155] Where, x i_max and x i_min Let x represent the maximum and minimum values ​​of the i-th data set. i * This represents the data after normalization; x i This indicates data that has not undergone normalization.

[0156] Analysis of the measured frequency domain transients revealed that, in all cases, the signal energy generated by transient distortion primarily exists above a certain frequency. Therefore, a Butterworth low-pass filter was used to isolate the per-unit data, yielding the main components of the system's inertial response. The Butterworth low-pass filter design process is as follows:

[0157] 1. Determine the performance specifications of the Butterworth low-pass filter:

[0158] Passband cutoff frequency Ω p Minimum passband attenuation αp Stopband start frequency Ω s and the maximum stopband attenuation α s .

[0159] 2. Calculate the normalized frequency:

[0160]

[0161]

[0162] Among them, Ω p Ω is the passband cutoff frequency. s λ is the stopband start frequency; p λ is the normalized passband cutoff frequency; s This is the normalized stopband start frequency.

[0163] When α = 3dB, Ω p =Ω c This refers to the cutoff frequency in the usual sense.

[0164] 3. Determine the Butterworth filter order N and Butterworth filter parameters C according to the design requirements:

[0165]

[0166]

[0167] in:

[0168]

[0169] Where, α p Minimum passband attenuation; α s is the maximum stopband attenuation; C is the Butterworth filter parameter; N is the Butterworth filter order.

[0170] Note when α p When the current is 3dB, C = 1;

[0171] 4. Look up the table using the N value:

[0172] The system function of the normalized Butterworth low-pass filter is obtained by looking up the N value in a table.

[0173] 5. After removing normalization, the low-pass filter H(s) is:

[0174]

[0175] Where s is the complex frequency variable in the Laplace transform; p is the complex frequency variable in the normalized Laplace transform.

[0176] The DFA algorithm can be used to remove existing "external fluctuations" in order to analyze the "normal" changes in signals. It ensures the detection of transient events in the inertial response under different scenarios. The steps of DFA detrending are as follows:

[0177] The original signal is integrated to remove the DC offset in the data, as shown in formula (9).

[0178]

[0179] Where y(k) is the integrated signal; k is the number of filtered signal samples; and x is the average value of the signal over the period from i=1 to k.

[0180] The integrated signal y(k) is then divided into multiple windows or boxes of variable length “n”. A least-squares first-order linear approximation is then performed on each window to represent the “trend” of that signal segment.

[0181] The trend of the integrated time series is removed by subtracting the local trend of each window from the integrated signal, as shown in Equation (10).

[0182] e(k)=y(k)-y n (k) (10)

[0183] Where e(k) is the detrending signal; y n (k) represents the trend of signals for each window.

[0184] The detrending signal e(k) is considered to be an approximation error. The root mean square fluctuation of equation (10) can be calculated using equation (11).

[0185]

[0186] Where F(n) is the root mean square of the detrending signal.

[0187] A larger F value indicates a greater transient response measured on the network. By comparing this response with other PMUs, the approximate location of the event can be determined. Simultaneously, setting an appropriate F value based on network characteristics determines the event's start time and assesses its suitability for inertia estimation.

[0188] After the above step S1, the preprocessed PUM data and events that can be used for inertia estimation are obtained.

[0189] In this embodiment, step S2 specifically includes:

[0190] Step S21: Analyze the operating characteristics of the synchronous generator and determine the physical model for estimating the inertia of the synchronous generator based on the state of the synchronous generator;

[0191] The inertia H of a synchronously operating unit is constant. Therefore, it can be assumed that the inertia of a synchronous generator is determined only by the operating state of the synchronous generator unit. When judging the inertia level, the key factor affecting the inertia, the "operating state", can be grasped to quickly and accurately estimate the system inertia, as shown in formula (12).

[0192]

[0193] Among them, H sync H is the equivalent inertia constant of the synchronous generator. Gk Let x be the inertia constant of the synchronous generator k. Gk Let n represent the start / stop status of generator k, and n be the number of synchronous generators.

[0194] Step S22: The characteristics of synchronous generator inertia and virtual inertia are different and need to be modeled and estimated separately. For virtual inertia estimation, a corresponding physical model is established based on the control means provided for virtual inertia.

[0195] In the electro-hydrogen fusion hub, synchronous generators provide instantaneous power exchange when frequency disturbances begin. In contrast, the virtual inertia response from the inverter interface source exhibits a certain time delay and different response characteristics. Furthermore, the properties of the virtual inertia output will change according to the topology transformation; therefore, the traditional method of using Δp, RoCof combined with the oscillation equation to explain its response characteristics at the beginning of the disturbance cannot accurately capture virtual inertia.

[0196] This invention represents and classifies the equivalent virtual inertia responses from different generator units at the system level. Currently, most topologies can be represented as variations in active power response based on RoCoF, frequency deviation, or a combination of both. Simultaneously, the responses of multiple virtual inertia topologies are aggregated.

[0197] The low-order system model based on droop control is shown in Equation (13).

[0198]

[0199] Among them, H droop (s) is the transfer function of the low-order system model with droop control, where T1 and T d1 H represents the time delay and time constant corresponding to droop control. droop Gain for droop control.

[0200] The virtual inertial response control block diagram based on droop control is shown below. Figure 2 As shown.

[0201] The low-order system model based on RoCoF control is shown in Equation (14).

[0202]

[0203] Among them, H RoCoF (s) is the transfer function of the low-order system model based on RoCoF control, T2 and T d2 To determine the time delay and time constant corresponding to RoCoF control, H RoCoF The gain is based on RoCoF control.

[0204] The control block diagram based on RoCoF virtual inertia response is as follows: Figure 3 As shown.

[0205] With the help of system identification, virtual inertia contribution TF Non-Sync The overall equivalent transfer function is:

[0206]

[0207] Among them, △P Non-Sync The power provided for the virtual inertia, Δf is the system frequency imbalance.

[0208] With all synchronous generators fully visible, the total power ΔP provided by the virtual inertial source can be calculated. Non-Sync Total power ΔP provided by synchronous generator Sync Separately, based on the individual control methods described above, they are combined through certain relationships to form the SISO inertia identification system. The overall virtual inertia transfer function is:

[0209]

[0210] Among them, H Non-Sync (s) is the transfer function of the system's virtual inertia, T d This is the equivalent time delay.

[0211] The control block diagram of the overall virtual inertia response is as follows: Figure 4 As shown.

[0212] Step S23: Based on the overall virtual inertia transfer function, identify system parameters to obtain virtual inertia, and combine it with the synchronous generator inertia to obtain system inertia.

[0213] Substitute the processed PMU data to calculate the total power imbalance and frequency imbalance of the system. Determine the imbalance borne by the synchronous generator by the magnitude of the synchronous generator's inertia, as shown in formula (17).

[0214]

[0215] Among them, △P Sync The imbalance borne by the synchronous generator; The rate of change of frequency imbalance;

[0216] The power provided by the virtual inertia source is separated from the total power imbalance, as shown in Equation (18).

[0217] ΔP Non-Sync =ΔP - ΔP Sync (18)

[0218] Where ΔP is the system power imbalance; ΔP Non-Syne Power provided for virtual inertia.

[0219] The power provided by the separated virtual inertia source, the system frequency imbalance, and the virtual equivalent inertia transfer function are substituted into the system identification toolbox of the mathematical software to obtain the virtual equivalent inertia. The system inertia is equal to the superposition of the synchronous generator inertia and the virtual inertia, as shown in formula (19).

[0220] H sys =H Non-Sync +H Sync (19)

[0221] Among them, H sys H represents the overall inertia of the system. Non-Sync This is the virtual equivalent inertia.

[0222] After step S2 above, the physical model for estimating the overall inertia is obtained.

[0223] In an embodiment, such as Figure 5 As shown, step S3 specifically includes:

[0224] Step S31: Select the input features to determine the number of neurons in the input layer of the extreme learning machine.

[0225] Feature quantities are the input data of the data model, and in principle, they are physical quantities that have a causal relationship with the output data. Considering that the object of calculation is the equivalent inertia of the power system, physical quantities that can directly affect the inertia level or the accuracy of the physical method estimation are selected. Physical quantities that have a greater impact on the calculated inertia are classified as primary input feature quantities, while physical quantities that have a smaller impact on the calculated inertia are classified as secondary feature quantities.

[0226] There are three main factors affecting system inertia: the inertial constant and grid connection status of the synchronous generator sets within the system, the inertial time constant and grid connection status of the virtual synchronous generators, and the total power of the gridded asynchronous motors. This is the objective existence of inertia and will not change due to differences in measurement data such as frequency. On the other hand, from a computational perspective, without considering the transient state at the time of disturbance, in the initial stage of the disturbance, only the disturbance power, static load voltage characteristics, and inertial support power act within the system, maintaining power balance. During this stage, the system's dynamic frequency response is entirely determined by the transfer and transformation between these three types of power. Therefore, theoretically, the system's inertia value can be uniquely determined simply by selecting the system frequency response, disturbance power, and load self-regulation power as input features. In addition, the location of the disturbance point, the location of the sampling point, the load level, and the proportion of various static loads will all affect the calculated value of inertia.

[0227] Based on the system composition, select the main input features and secondary features from the variables considered above.

[0228] Step S32: Establish the basic structural framework of the Extreme Learning Machine and use the ant colony algorithm to optimize the weights and thresholds of the input layer and hidden layer of the Extreme Learning Machine.

[0229] For a specific training sample set of N sets of data (x p ,t p ), x p =[x p1 ,x p2 ,…,x pn ] T ∈R n The input vector includes the system's equivalent inertia estimation results from the physical model, the location of the perturbation point, the location of the sampling point, frequency data, power data, etc.; t p =[t p1 ,t p2 ,…,t pm ] T ∈R m This is the output vector, including the corrected equivalent inertia calculation results. The superscript T denotes the matrix transpose; R n R is a vector containing n real numbers; m Let m be a vector containing m real numbers.

[0230] The mathematical model of the Extreme Learning Machine can be expressed as:

[0231]

[0232] Where M is the number of hidden layer units; g(x) is the activation function; w q =[w q1 ,w q2 ,…,w qn] T β is the weight vector connecting the input vector and the q-th hidden layer unit; q =[β q1 ,β q2 ,…,β qn ] T b is the weight vector connecting the q-th hidden layer unit and the output vector; q The bias is set for the q-th hidden layer unit; o j =[o j1 ,o j2 ,…,o jm ] T Here is the predicted output vector; j is the j-th data set; N is the number of data sets.

[0233] If the prediction result is o j It can accurately approximate the actual results t of these N sets of samples j Then the following equation holds:

[0234]

[0235] Among them, o j For the prediction result, t j For the actual result; ||o j -t j || represents the norm of the difference in the results;

[0236] Substitute formula (21) into formula (20):

[0237]

[0238] Representing formula (22) in matrix form, we get:

[0239] Hβ=T (23)

[0240] in:

[0241]

[0242]

[0243]

[0244] Where H is the hidden layer output matrix of the neural network; β is the output weight matrix, which is an important parameter for establishing the error fitting model; and T is the prediction result of the frequency situation characteristics.

[0245] In the Extreme Learning Machine algorithm, w is determined randomly. i and b iThe hidden layer output matrix H is then uniquely determined. Therefore, this problem can be transformed into solving a linear system problem Hβ=T. The output weight β can then be approximately obtained by the following equation:

[0246]

[0247] in, Let H be the generalized inverse matrix of H.

[0248] After establishing the basic framework of the Extreme Learning Machine, it is optimized and trained based on the ant colony algorithm. The specific steps are as follows:

[0249] Step 1. Determine the topology of the Extreme Learning Machine, namely the number of neurons in the input layer, the number of neurons in the hidden layer, and the number of neurons in the output layer;

[0250] Step 2. Encode the weights and thresholds from the input layer to the hidden layer in the extreme learning machine to obtain the initial population;

[0251] Step 3. Decode the weights and thresholds, and feed them into the training network of the Extreme Learning Machine. Use the training samples to train the machine.

[0252] Step 4. After training is complete, test the sample and use the sum of squared errors between the expected value and the predicted value of the test sample as the fitness function.

[0253] Step 5. Select, crossover, and mutate the population to obtain a new population. If the conditions are met, the network weights and thresholds with the minimum sum of squared errors are obtained. If the conditions are not met, return to step 2.

[0254] Step 6. Input the optimized weights and thresholds into the training network, calculate the hidden layer output matrix H, and solve for the generalized inverse matrix of H.

[0255] Step 7. Calculate the output layer weights using formula (27);

[0256] Step 8. Feed the test samples into the Extreme Learning Machine model for prediction.

[0257] Step S33: Connect the physical model and the data model in sequence to obtain the data-physical fusion inertia estimation model, establish an evaluation standard model, and judge the accuracy of the data-physical fusion inertia estimation model.

[0258] The estimation results obtained from the physical model are used as input to the data model in a serial manner. Then, the data model is configured, trained, and tested using the ant colony algorithm to obtain the inertia correction model. The accuracy of the estimation results is judged using the following evaluation model, with the following evaluation metrics:

[0259] Root Mean Square Error (RMSE):

[0260]

[0261] Among them, H r,l The actual value of inertia in the test data; H p,l The inertia prediction value is for the test data.

[0262] When the root mean square error meets expectations, the inertia assessment model can be applied to real-time inertia estimation.

[0263] like Figure 6 As shown, the inertia assessment device for an electric hydrogen energy hub according to the present invention includes:

[0264] The data processing module 41 is used to read the raw data measured by the synchronous phasor measurement device, and to standardize the raw data to make the data dimensionless for easy calculation. Then, filtering technology is used to eliminate problems such as oscillation and noise in the raw data, and the events that can be used for inertia estimation are determined by detrending. The obtained data serves as the data basis for subsequent modules to reduce estimation errors.

[0265] The physical model inertia estimation module 42, based on the synchronous generator inertia characteristics that are only related to the start-up state and the virtual inertia characteristics that are controlled based on RoCoF, frequency deviation, or a combination of both, finds the relationship between the input quantity and inertia, constructs a flowchart, and calculates the transfer function. Based on the preprocessed data, it performs preliminary synchronous inertia and virtual inertia estimation, and superimposes them to obtain the total inertia.

[0266] The data model inertia estimation module 43 determines the number of neurons in the input layer of the extreme learning machine, and uses the ant colony algorithm to optimize the weights and thresholds of the input layer and hidden layer of the extreme learning machine to obtain the extreme learning machine model. The inertia initially estimated by the physical model is substituted into the data model for correction to obtain the total inertia of the electric hydrogen energy hub.

[0267] In summary, this invention discloses a data-physical fusion-based inertia assessment method for an electric-hydrogen energy hub. In scenarios of increasing new energy penetration and the integration of power systems with hydrogen energy, this invention addresses the issue of low accuracy associated with single physical models by proposing an inertia estimation method based on data-physical fusion. This invention uses data obtained from PMU monitoring as the data foundation and identifies inertia centers for data preprocessing, during which events suitable for inertia estimation are determined. Next, a physical model for inertia estimation is established based on the synchronous generator's state. This model is then combined with a physical model established using virtual inertia support methods, and the two physical models are superimposed to obtain a physical model for calculating the overall system inertia. To further improve the system's accuracy, an extreme learning machine data model is introduced, fusing the physical model and the data model to obtain the data-physical fusion-based inertia estimation method.

[0268] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0269] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0270] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0271] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0272] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0273] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for evaluating the inertia of an electric hydrogen energy hub, characterized in that, Includes the following steps: The center of inertia is determined, and the data measured by the synchronous phasor unit of the center of inertia is used as the data basis. The measured data is then normalized, filtered, and detrended preprocessed. A physical model of the synchronous generator inertia is established based on the characteristics of synchronous generator inertia and the virtual inertia control method. Preprocessed data is substituted into the physical model to estimate the synchronous and asynchronous inertia respectively. These are then superimposed to obtain the preliminary overall inertia of the electric-hydrogen energy hub, including: Analyze the operating characteristics of the synchronous generator and determine the physical model for estimating the generator's inertia based on its state: (12) Among them, H sync H is the equivalent inertia constant of the synchronous generator. Gk Let x be the inertia constant of the k-th synchronous generator. Gk This represents the start / stop status of generator k, where n is the number of synchronous generators; A corresponding physical model is established based on the control methods provided for virtual inertia, which is used for virtual inertia estimation; For the low-order system model based on droop control, as shown in Equation (13): (13) Among them, H droop (s) is the transfer function of the low-order droop control system model, where T1 and T d1 H represents the time delay and time constant corresponding to droop control. droop Gain for droop control; For the low-order system model based on RoCoF control, as shown in equation (14): (14) Among them, H RoCoF (s) is the transfer function of the low-order system model based on RoCoF control, T2 and T d2 To determine the time delay and time constant corresponding to RoCoF control, H RoCoF Gain based on RoCoF control; With the help of system identification, virtual inertia contribution TF Non-Sync The overall equivalent transfer function is: (15) Among them, △P Non-Sync The power provided for the virtual inertia, Δf is the system frequency imbalance; The overall virtual inertia transfer function is: (16) Among them, H Non-Sync (s) is the transfer function of the system's virtual inertia, T d For equivalent time delay; The virtual inertia is obtained by identifying system parameters based on the overall virtual inertia transfer function, and then combined with the synchronous generator inertia to obtain the system inertia: Substitute the processed PMU data to calculate the total power imbalance and frequency imbalance of the system. Determine the imbalance borne by the synchronous generator by the magnitude of the synchronous generator's inertia, as shown in formula (17): (17) Among them, △P Sync The imbalance borne by the synchronous generator; The rate of change of frequency imbalance; The power provided by the virtual inertia source is separated from the total power imbalance, as shown in Equation (18): (18) Where ΔP is the system power imbalance; The virtual equivalent inertia is obtained based on the power provided by the separated virtual inertia source, the system frequency imbalance, and the virtual equivalent inertia transfer function. The system inertia is equal to the superposition of the synchronous generator inertia and the virtual inertia, as shown in formula (19): (19) Among them, H sys H represents the overall inertia of the system. Non-Sync For virtual equivalent inertia; A data model for inertia assessment is established for the Extreme Learning Machine (ELM). After determining the framework of the ELM, the data model is trained using the ant colony algorithm to obtain the optimal parameters. The physical model and data model of the synchronous generator inertia are combined in a serial manner to form a data-physical fusion model. The total inertia obtained from the physical model is input into the data model for correction to obtain the total inertia of the hydrogen-electric energy hub.

2. The method for evaluating the inertia of an electric hydrogen energy hub according to claim 1, characterized in that, The process of determining the center of inertia, using data measured by the synchronous phasor unit of the center of inertia as the data basis, and performing preprocessing such as standardization, filtering, and detrending on the measured data includes: By introducing the center of inertia to select the optimal frequency sampling point, and using formula (1) to traverse each node in the system, the optimal frequency sampling point is found. E The node with the smallest value is used as the frequency sampling point. (1) Among them, F E The cost function value; u represents the node number; i represents the generator number; H i S i Let be the inertial constant and rated capacity of generator i, respectively; d 2 (u,i) represents the squared distance between node u and generator node i; u∈m represents the u-th node in the system node set; Based on the data obtained by the synchronous phasor measurement device at the optimal frequency sampling point, the data is normalized and mapped to the [0,1] interval. The normalization method is shown in formula (2): (2) Where, x i_max and x i_min This represents the maximum and minimum values ​​of the i-th data set. This represents the data after normalization; x i This represents data that has not undergone normalization. The components of the system inertial response are obtained by isolating the normalized data using a Butterworth low-pass filter. Perform DFA detrending.

3. The method for evaluating the inertia of an electric hydrogen energy hub according to claim 2, characterized in that, The DFA detrending includes: The original signal is integrated to remove the DC offset present in the data; The integrated signal is divided into multiple windows or boxes of variable length n, and then a least-squares first-order linear approximation is performed on each window to represent the "trend" of the signal segment. Calculate the root mean square (RMS) of the detrending signal, determine the location of the event based on the RMS, set an appropriate RMS based on the network characteristics, determine the start time of the event, assess the applicability of the event for inertia estimation, and complete the DFA detrending.

4. The method for evaluating the inertia of an electric hydrogen energy hub according to claim 1, characterized in that, The aforementioned method establishes a data model for inertia evaluation using an extreme learning machine. After determining the framework of the extreme learning machine, the data model is trained using an ant colony algorithm to obtain optimal parameters. The physical model and data model of the synchronous generator inertia are then combined in a serial manner to form a data-physical fusion model. The total inertia obtained from the physical model is input into the data model for correction, resulting in the total inertia of the hydrogen-electric energy hub, which includes: Select the input features to determine the number of neurons in the input layer of the extreme learning machine; A basic structural framework for the Extreme Learning Machine (ELM) is established, and the ant colony algorithm is used to optimize the weights and thresholds of the input layer and hidden layer of the ELM. By sequentially connecting the physical model and the data model, a data-physical fusion inertia estimation model is obtained. An evaluation standard model is then established to determine the accuracy of the data-physical fusion inertia estimation model. The accuracy of the estimation results is judged using the following evaluation model, with the following evaluation indicators: Root Mean Square Error (RMSE): (28) Among them, H r,l The actual value of inertia in the test data; H p,l The inertia prediction value for the test data; When the root mean square error meets expectations, the data-physical fusion inertia estimation model can be applied to real-time inertia estimation.

5. The method for evaluating the inertia of an electric hydrogen energy hub according to claim 4, characterized in that, The establishment of the basic structural framework of the Extreme Learning Machine and the optimization of the weights and thresholds of the input layer and hidden layer of the Extreme Learning Machine using the ant colony algorithm include: For a specific training sample set of N sets of data (x p ,t p ), x p =[x p1 ,x p2 ,…,x pn ] T ∈R n The input vector includes the system's equivalent inertia estimation results from the physical model, the location of the perturbation point, the location of the sampling point, frequency data, and power data; t p =[t p1 ,t p2 ,…,t pm ] T ∈R m The output vector includes the corrected equivalent inertia calculation results; The mathematical model of the Extreme Learning Machine is expressed as: (20) Where M is the number of hidden layer units; g(x) is the activation function; w q =[w q1 ,w q2 ,…,w qn ] T β is the weight vector connecting the input vector and the q-th hidden layer unit; q =[β q1 , β q2 ,…, β qn ] T b is the weight vector connecting the q-th hidden layer unit and the output vector; q The bias is set for the q-th hidden layer unit; o j =[o j1 ,o j2 ,…,o jm ] T is the predicted output vector; j is the j-th data set; N is the total number of data sets; the superscript T indicates the transpose of the matrix; If the prediction result is o j It can accurately approximate the actual results t of these N sets of samples j Then the following equation holds: (21) Among them, o j For the prediction result, t j For actual results; The norm of the difference in the results; Substitute formula (21) into formula (20): (22) Equation (22) can be represented as a matrix: (23) in: (24) (25) (26) Where H is the hidden layer output matrix of the neural network; β is the output weight matrix, which is an important parameter for establishing the error fitting model; and T is the prediction result of the frequency situation characteristics. In the Extreme Learning Machine algorithm, w is determined randomly. i and b i The hidden layer output matrix H is then uniquely determined; according to the linear system problem Hβ=T, the output weight β is approximately obtained by the following formula: (27) Among them, H † Let H be the generalized inverse matrix of H.

6. The method for evaluating the inertia of an electric hydrogen energy hub according to claim 5, characterized in that, The establishment of the basic structural framework of the Extreme Learning Machine (ELM) and the optimization of the weights and thresholds of the input and hidden layers using the ant colony algorithm include: after establishing the basic framework of the ELM, optimizing and training it based on the ant colony algorithm, including: Step 1. Determine the topology of the Extreme Learning Machine, namely the number of neurons in the input layer, the number of neurons in the hidden layer, and the number of neurons in the output layer; Step 2. Encode the weights and thresholds from the input layer to the hidden layer in the extreme learning machine to obtain the initial population; Step 3. Decode the weights and thresholds, and feed them into the training network of the Extreme Learning Machine. Use the training samples to train the machine. Step 4. After training is complete, test the sample and use the sum of squared errors between the expected value and the predicted value of the test sample as the fitness function. Step 5. Select, crossover, and mutate the population to obtain a new population. If the conditions are met, the network weights and thresholds with the minimum sum of squared errors are obtained. If the conditions are not met, return to step 2. Step 6. Input the optimized weights and thresholds into the training network, calculate the hidden layer output matrix H, and solve for the generalized inverse matrix Hi. † ; Step 7. Calculate the output layer weights using formula (27); Step 8. Feed the test samples into the Extreme Learning Machine model for prediction.

7. A device for evaluating the inertia of an electric hydrogen energy hub, characterized in that, include: The data processing module is used to read the raw data measured by the synchronous phasor measurement device, and to normalize the raw data to make it dimensionless for easy calculation. Then, filtering technology is used to eliminate problems such as oscillation and noise in the raw data, and the events that can be used for inertia estimation are determined by detrending. The obtained data serves as the data basis for the subsequent modules to reduce estimation errors. The physical model inertia estimation module, based on the synchronous generator inertia characteristics relevant only to the start-up state and the virtual inertia characteristics controlled based on RoCoF, frequency deviation, or a combination of both, finds the relationship between the input quantity and inertia, constructs a flowchart, and calculates the transfer function. It then performs preliminary synchronous and virtual inertia estimations based on preprocessed data, and superimposes these estimates to obtain the total inertia, including: Analyze the operating characteristics of the synchronous generator and determine the physical model for estimating the generator's inertia based on its state: (12) Among them, H sync H is the equivalent inertia constant of the synchronous generator. Gk Let x be the inertia constant of the k-th synchronous generator. Gk This represents the start / stop status of generator k, where n is the number of synchronous generators; A corresponding physical model is established based on the control methods provided for virtual inertia, which is used for virtual inertia estimation; For the low-order system model based on droop control, as shown in Equation (13): (13) Among them, H droop (s) is the transfer function of the low-order droop control system model, where T1 and T d1 H represents the time delay and time constant corresponding to droop control. droop Gain for droop control; For the low-order system model based on RoCoF control, as shown in equation (14): (14) Among them, H RoCoF (s) is the transfer function of the low-order system model based on RoCoF control, T2 and T d2 To determine the time delay and time constant corresponding to RoCoF control, H RoCoF Gain based on RoCoF control; With the help of system identification, virtual inertia contribution TF Non-Sync The overall equivalent transfer function is: (15) Among them, △P Non-Sync The power provided for the virtual inertia, Δf is the system frequency imbalance; The overall virtual inertia transfer function is: (16) Among them, H Non-Sync (s) is the transfer function of the system's virtual inertia, T d For equivalent time delay; The virtual inertia is obtained by identifying system parameters based on the overall virtual inertia transfer function, and then combined with the synchronous generator inertia to obtain the system inertia: Substitute the processed PMU data to calculate the total power imbalance and frequency imbalance of the system. Determine the imbalance borne by the synchronous generator by the magnitude of the synchronous generator's inertia, as shown in formula (17): (17) Among them, △P Sync The imbalance borne by the synchronous generator; The rate of change of frequency imbalance; The power provided by the virtual inertia source is separated from the total power imbalance, as shown in Equation (18): (18) Where ΔP is the system power imbalance; The virtual equivalent inertia is obtained based on the power provided by the separated virtual inertia source, the system frequency imbalance, and the virtual equivalent inertia transfer function. The system inertia is equal to the superposition of the synchronous generator inertia and the virtual inertia, as shown in formula (19): (19) Among them, H sys H represents the overall inertia of the system. Non-Sync For virtual equivalent inertia; The data model inertia estimation module determines the number of neurons in the input layer of the Extreme Learning Machine (ELM) and uses the ant colony algorithm to optimize the weights and thresholds of the input and hidden layers of the ELM to obtain the ELM model. The inertia initially estimated by the physical model is substituted into the data model for correction to obtain the overall inertia of the hydrogen-electric energy hub.

8. The inertia assessment device for an electric hydrogen energy hub according to claim 7, characterized in that, In the data processing module, the optimal frequency sampling point is selected by using the inertia center. Formula (1) is used to iterate through each node in the system to find the F frequency sampling point. E The node with the smallest value is used as the frequency sampling point. (1) Among them, F E The cost function value; u represents the node number; i represents the generator number; H i S i Let be the inertial constant and rated capacity of generator i, respectively; d 2 (u,i) represents the squared distance between node u and generator node i; u∈m represents the u-th node in the system node set; Based on the data obtained by the synchronous phasor measurement device at the optimal frequency sampling point, the data is normalized and mapped to the [0,1] interval. The normalization method is shown in formula (2): (2) Where, x i_max and x i_min This represents the maximum and minimum values ​​of the i-th data set. This represents the data after normalization; x i This represents data that has not undergone normalization. The components of the system inertial response are obtained by isolating the normalized data using a Butterworth low-pass filter. Perform DFA detrending.

Images