A power system inertia shortage identification method based on frequency feature extraction

Abstract

The application discloses a kind of inertia shortage identification methods of power system based on frequency feature extraction, belong to novel power system planning technical field.The method includes: the system frequency swing equation considering new energy network type virtual inertia response delay is constructed, and the key frequency features for identifying inertia level shortage are extracted;Inertia level shortage identification model is constructed using deep learning algorithm;Based on historical unit commitment data, the identification model is trained offline;Unit commitment optimization model is constructed, and the day-ahead unit commitment scheme is obtained based on day-ahead new energy output and load prediction curve, and the trained identification model is used to verify the inertia level of the scheme.The application uses system steady-state operation data to post-check day-ahead unit commitment scheme, can accurately identify the inertia level shortage of system after high proportion of new energy access, ensure the frequency safety of novel power system operation.

Description

Technical Field

[0001] This invention relates to the field of novel power system planning technology, and in particular to a method for identifying power system inertia shortage based on frequency feature extraction. Background Technology

[0002] Inertia is fundamental for new power systems to resist high-power disturbances and maintain frequency security. However, with the increasing proportion of renewable energy, the trend of decreasing synchronous inertia levels in power systems is becoming increasingly apparent. Therefore, renewable energy inertia support technology based on virtual inertial control has been implemented. The heterogeneity of frequency support capabilities between synchronous machines and renewable energy generation presents new challenges to frequency security in the operation of new power systems. Furthermore, the virtual inertia support and frequency regulation capabilities of renewable energy are affected by grid connection methods and weather changes, making it difficult to accurately identify potential inertia shortages in the system. Therefore, accurately identifying inertia shortages after a high proportion of renewable energy is integrated into the grid is crucial for analyzing the frequency security and reliable operation of new power systems.

[0003] Therefore, there is an urgent need for a method that can accurately and quickly identify the inertia level shortage of new power systems, so as to provide verification and guarantee for the safe operation of the system. Summary of the Invention

[0004] The purpose of this invention is to provide a power system inertia shortage identification method based on frequency feature extraction, which solves the problem of difficulty in accurately identifying system inertia level shortage caused by the access of new energy sources in the prior art, provides a reference for subsequent inertia level supplementation planning, and can ensure the frequency security of the operation of new power systems.

[0005] To achieve the above objectives, this invention provides a method for identifying power system inertia shortage based on frequency feature extraction, comprising the following steps: S1. Construct a system frequency swing equation that considers the virtual inertia response delay of new energy grid-type systems, so as to extract key frequency features that can accurately identify the system inertia level shortage. S2. Using deep learning algorithms, a new identification model for power system inertia level shortage is constructed to achieve online identification of system inertia level shortage; S3. The identification model is trained offline using historical unit combination data to improve the reliability and speed of the model identification results; S4. Construct a new power system unit combination optimization model. Based on the day-ahead renewable energy output and load forecast curves, provide a day-ahead unit combination scheme, and then use the trained identification model to verify the inertia level of the scheme.

[0006] Preferably, in step S1, the system frequency swing equation considering the virtual inertia response delay of the new energy network is expressed as: ; In the formula, The first-order response delay for virtual inertia in new energy network structures; , These are the system's synchronous inertia energy and virtual inertia energy, respectively. , These are the primary frequency regulation power and the unbalanced power of the system, respectively. The damping coefficient of the system; The system's rated frequency; This refers to the frequency change of the system. This represents the time elapsed after the disturbance occurred.

[0007] Preferably, in step S1, the key frequency features extracted to identify the inertia level shortage of the identification system are: ; In the formula, The time when the disturbance occurs. This represents the maximum rate of frequency change that the power system can withstand. If any of the above key frequency characteristics are met, it indicates a shortage of system inertia, requiring timely replenishment of inertia resources.

[0008] Preferably, in step S2, a multi-layer neural network structure with convolutional computing capability and data feature dimensionality reduction capability is used to build the main network, which is then connected in series with a network layer with data shape compression capability and a network layer with multi-neuron connection capability, and finally connected in series with a functional module with key frequency feature judgment capability to form the recognition model.

[0009] Preferably, step S3 includes the following sub-steps: S31. Obtain historical unit combination data, filter the input data types required for training, and form offline training dataset input data; S32. Based on historical unit combination data, set different disturbance powers in transient simulation, obtain the system frequency change curves corresponding to each operating moment in batches, extract the system frequency change curves, obtain the maximum frequency change rate of the system in batches, and form offline training dataset output data. S33. Combine the input and output data of the offline training dataset to form the offline training dataset; divide the offline training dataset into training set and test set according to the proportion; S34. Set the optimizer, training batch, learning rate, number of training iterations and safety loss function during the training process. Based on the offline training test set performance of the recognition model, obtain the optimal inertia level shortage prediction and recognition model when the set stopping conditions are met.

[0010] Preferably, in step S34, the security loss function is expressed as: ; In the formula, This represents the loss function of the neural network in the recognition model. This indicates the recognition model in terms of network hyperparameters. The loss function used to predict the maximum rate of change of the system frequency is as follows.

[0011] Preferably, in step S4, the unit combination optimization model is as follows: under the constraints of the new power system operation, with the minimum operating cost as the objective function, the starting mode of each traditional unit, the output power of each traditional unit, and the actual output power of each new energy power station are obtained in each time period within the scheduling cycle; wherein, the objective function L is expressed as: ; In the formula, This represents the total cost of the output power of a synchronous generator unit. This represents the total cost of starting and stopping a synchronous generator unit. This represents the total cost of penalties for absorbing new energy power plants.

[0012] Preferably, the total cost of the output power of the synchronous generator unit Total start-up and shutdown costs of synchronous generator units Total cost of absorbing renewable energy power plants They are represented as follows: ; ; ; In the formula, Indicates the total number of time periods. This represents the total number of traditional generating units in the new power system; , , These represent traditional generating units. First, second, and third output power cost coefficients; , These represent traditional generating units. The startup and shutdown cost coefficients, Indicates traditional units The binary start / stop status; This represents the total number of new energy power stations in the new power system; Indicates new energy power station The cost coefficient of the disposal penalty; , They represent new energy power stations. The predicted output power and the actual output power.

[0013] Preferably, in step S4, the inertia shortage of the system is identified online through the following sub-steps: S41. Obtain the day-ahead renewable energy output and load forecast curves, and obtain the day-ahead unit combination scheme through the unit combination optimization model; S42. Filter the input data features in the day-ahead unit combination scheme to form an input data sequence and input it into the identification model; S43. The identification model verifies the inertia level of the scheme by quickly predicting and judging key frequency features.

[0014] Preferably, the system for implementing the above method includes: The system frequency swing equation module is used to reflect the network-type virtual inertia response delay provided by new energy sources in the new power system, and to extract key frequency features that can accurately identify system inertia shortages. The inertia shortage identification module is used to build an inertia level shortage identification model for a new type of power system and train it offline to achieve online identification of system inertia level shortage. The day-ahead unit combination optimization module is used to provide day-ahead unit combination schemes based on the day-ahead renewable energy output and load forecast curves, and then use the identification model to verify the inertia level of the scheme. The system frequency swing equation module is used to reflect the network-type virtual inertia response delay provided by new energy sources in the new power system, and to extract key frequency features that can accurately identify system inertia shortages. The inertia shortage identification module is used to build an inertia level shortage identification model for a new type of power system and train it offline to achieve online identification of system inertia level shortage. The day-ahead unit combination optimization module is used to provide a day-ahead unit combination scheme based on the day-ahead renewable energy output and load forecast curves, and then use the identification model to verify the inertia level of the scheme.

[0015] Therefore, the power system inertia shortage identification method based on frequency feature extraction using the above structure has the following beneficial effects: (1) This invention constructs a system frequency swing equation that considers the response delay of virtual inertia in new energy grids. It can more accurately describe the frequency dynamic response process of new power systems after disturbances, and extract key frequency features to identify the shortage of system inertia level. It effectively solves the problem that the virtual inertia support and frequency regulation capability of new energy are affected by grid connection mode and weather changes, making it difficult to accurately identify the hidden danger of inertia shortage.

[0016] (2) This invention uses deep learning algorithms to construct a system inertia level shortage identification model, and performs offline training on the model based on historical unit combination data. The system's steady-state operation data can be used to realize rapid post-verification of the inertia level of the day-ahead unit combination scheme, which improves the reliability and speed of online identification.

[0017] (3) The present invention constructs a new type of power system unit combination optimization model, and gives a day-ahead unit combination scheme based on the day-ahead new energy output and load forecast curve. Then, the proposed identification model is used to verify the inertia level of the scheme on the hour-level time scale, which can ensure the frequency security of the new power system operation.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] Figure 1 This is a flowchart of a power system inertia shortage identification method based on frequency feature extraction according to the present invention; Figure 2 This is a schematic diagram of the improved North American NPCC-140 node system in an embodiment of the present invention; Figure 3 This is an example of the error distribution in the prediction of key frequency change features by the identification model after offline training in this embodiment of the invention; Figure 4 The above are typical daily wind power output and load forecast curves in an embodiment of the present invention. Figure 5 This is the prediction result of the trained recognition model on a typical day for the key frequency change characteristics of the unit combination scheme in an embodiment of the present invention. Figure 6 This is an example of how the identification model identifies periods of inertia shortage in the unit combination scheme on a typical day in an embodiment of the present invention. Detailed Implementation

[0020] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0022] Example 1 like Figure 1 As shown, this embodiment of the invention provides a method for identifying power system inertia shortage based on frequency feature extraction, including the following steps: S1. Construct a system frequency swing equation that considers the virtual inertia response delay of new energy grid-type systems in order to extract key frequency features that can identify the shortage of system inertia level.

[0023] New energy power generation employs virtual synchronous machine technology, connecting to the grid as a voltage source by simulating the electromagnetic response principle of synchronous generator sets. It participates in system frequency regulation and provides virtual inertia through the implementation of corresponding control strategies. Unlike synchronous generator sets, where inertia is an inherent physical characteristic, new energy generator sets connected to the grid using virtual synchronous machine technology introduce frequency response delays due to power measurement, signal generation, and transmission. In this invention, considering the time delay present in the virtual inertia response, the frequency change process of the novel power system after disturbance can be represented by an equivalent swing equation: ; In the formula, The first-order response delay for virtual inertia in new energy network structures; , These are the system's synchronous inertia energy and virtual inertia energy, respectively. , These are the primary frequency regulation power and the unbalanced power of the system, respectively. The damping coefficient of the system; The system's rated frequency; This refers to the frequency change of the system. This represents the time elapsed after the disturbance occurred.

[0024] According to the frequency regulation process of the new power system, inertia instantaneously provides inertial energy to resist frequency changes at the initial moment after the system is disturbed. Because of the inherent dead time of the governor's startup during primary frequency regulation, the maximum rate of frequency change that occurs at the instant of system disturbance can fully reflect the shortage of system inertia. By combining the frequency changes after system disturbance with the set critical frequency characteristic safety range, the shortage of system inertia can be judged through the system's critical frequency change characteristics.

[0025] In this invention, the key frequency change features extracted from the system frequency swing equation are represented as follows: ; In the formula, The time when the disturbance occurs. This represents the maximum rate of frequency variation that the power system can withstand. If any of the above key frequency characteristics are met, it indicates a shortage of system inertia, causing the system frequency variation to exceed the safety threshold, requiring timely replenishment of inertia resources.

[0026] S2. Using deep learning algorithms, a new identification model for power system inertia level shortage is constructed.

[0027] First, a multi-layer neural network structure with convolutional computing capability and data feature dimensionality reduction capability is used to build the main network. Then, it is connected in series with a network layer with data shape compression capability and a network layer with multi-neuron connection capability. Finally, it is connected in series with a functional module with key frequency feature judgment capability to form a recognition model.

[0028] In implementation, the structure of the novel power system inertia level shortage identification model is as follows: input data layer, first processing layer, second processing layer, third processing layer, fully connected layer, output data layer, and judgment module. Each processing layer includes a convolutional layer, a batch normalization layer, and a pooling layer connected in sequence. The model input is a one-dimensional data sequence formed by filtering historical unit combination schemes, and the model output is the maximum frequency change rate of the system. The processing layer is used to extract data features, normalize and reduce the dimensionality of the data, and the fully connected layer is used to organize and compress the extracted data features.

[0029] Convolutional layers perform convolution calculations on the input data at different scales to capture potential mechanistic relationships and data features. Batch normalization layers normalize input data with different numerical ranges to improve the training efficiency and accuracy of neural networks. Pooling layers reduce the dimensionality of the normalized data while preserving the original features. Fully connected layers connect the input and output data layers, implementing the mechanisms captured by the network and establishing a feature mapping relationship between steady-state input data and transient output data. In convolutional layers, a large number of convolutional kernels slide along a single direction of the sequence data. Each process only convolves with a portion of the sequence and generates local feature mappings, which are expressed through activation functions. Subsequently, pooling layers compress the local features into a fixed-length vector to reduce computational complexity. Through multiple combinations of convolutional and pooling layers, deeper data features are gradually discovered. These features are then flattened into feature vectors and input to the fully connected layer, finally mapping to the output space to form the output data.

[0030] Specifically, the number of output channels of the convolutional layers in the first, second, and third processing layers are 32, 64, and 128, respectively, the kernel length is 3, they all use linear activation functions, they all adopt the same padding strategy, and the convolution stride is set to 1; the pooling window size of the pooling layer is 2; the number of neurons in the fully connected layer is 64, the activation function is ReLU, and L2 regularization is used to normalize the data.

[0031] The functional module with the ability to determine key frequency features realizes the inequality judgment of the above-mentioned key frequency features through logical discrimination.

[0032] S3. Using historical unit combination data, train the constructed identification model offline, including the following sub-steps: S31. Obtain historical unit combination data, filter the input data types required for training, and form offline training dataset input data.

[0033] The required data types selected from historical unit combination data include: the start-up mode of each synchronous unit, the output of each synchronous unit, the actual output of each new energy unit, the virtual inertia constant of each new energy generator unit, the inertia constant of each synchronous unit, and the primary frequency regulation rate of each governor. Through data processing, the input data types required for offline training of the recognition model are formed, including: the total system inertia level, the system synchronous inertia level, the system virtual inertia level, and the total primary frequency regulation rate of the governor within each time period of the scheduling cycle.

[0034] The system synchronous inertia level, system virtual inertia level, system total inertia level, and total primary frequency regulation rate of the speed controller for each time period are expressed as follows: ; ; ; ; In the formula, , , , These represent the time periods. The system's synchronous inertia level, virtual inertia level, total inertia level, and total primary frequency regulation rate of the speed controller are all considered. Indicates the first The inertia constant of the synchronous generator unit Indicates the first Taiwan synchronous generator unit during the time period The binary start / stop status within. Indicates the first Taiwan synchronous generator unit during the time period Internal output power, Indicates the first A new energy unit during the period The virtual inertia constant within, Indicates the first A new energy unit during the period The actual output power inside, Indicates the first The speed governor of the synchronous generator unit during the time period The frequency regulation rate within the range. The scheduling cycle of the scheduling plan is 24 hours.

[0035] S32. Based on historical unit combination data, set different disturbance powers in transient simulation and obtain the system frequency change curves corresponding to each operating moment in batches. Extract the system frequency change curves, obtain the maximum frequency change rate of the system in batches, and form offline training dataset output data.

[0036] By setting the magnitude and location of active power disturbances in different scenarios within the power system simulation software, frequency variation curves for each scheduling plan are obtained based on transient simulation calculations, and thus the maximum frequency variation rate for each scheduling plan is derived. The disturbance magnitude ranges from 5% to 10% of the load at each node within the system.

[0037] S33. Combine the input and output data of the offline training dataset to form the offline training dataset, and divide the offline training dataset into a training set and a test set according to the following ratio: 70% training set and 30% test set.

[0038] S34. Set the optimizer, training batch, learning rate, number of training iterations, and safety loss function during the training process. Based on the offline training test set performance of the recognition model, obtain the optimal inertia level shortage prediction and recognition model when the set stopping conditions are met.

[0039] The ADAM optimizer was selected, with a training batch size of 32, a learning rate of 1e-5, and 180 training iterations. The stopping condition was that the number of training iterations reached the set number; one iteration was defined as completing the training of the recognition model based on the partitioned sample set.

[0040] In each training batch, the frequency feature prediction model is trained in the following way: the sample dataset of the current batch is input into the recognition model for training, and the neural network hyperparameters in the recognition model are adjusted by backpropagation based on the learning rate and optimizer to minimize the security loss function.

[0041] The security loss function is expressed as: ; In the formula, This represents the loss function of the neural network in the recognition model. This indicates the recognition model in terms of network hyperparameters. The loss function used to predict the maximum rate of change of the system's frequency is described below. By applying the offline training safety loss function of the neural network in the identification model, the prediction of the maximum rate of change of the system's frequency by the identification model can be made slightly more severe than the actual value, thus making the identification result more reliable and secure.

[0042] S4. Construct a new power system unit combination optimization model. Based on the day-ahead renewable energy output and load forecast curves, provide day-ahead unit combination schemes, and then use the identification model to verify the inertia level of the schemes.

[0043] The constructed unit combination optimization model, under the operational constraints of the new power system, uses minimum operating cost as the objective function to obtain the start-up mode of each traditional unit, the output power of each traditional unit, and the actual output power of each new energy power station during each time period within the dispatch cycle; where the objective function L is expressed as: ; In the formula, This represents the total cost of the output power of a synchronous generator unit. This represents the total cost of starting and stopping a synchronous generator unit. This represents the total cost of penalties for absorbing new energy power plants.

[0044] Total cost of the output power of the synchronous generator unit Total start-up and shutdown costs of synchronous generator units Total cost of absorbing renewable energy power plants They are represented as follows: ; ; ; In the formula, Indicates the total number of time periods. This represents the total number of traditional generating units in the new power system; , , These represent traditional generating units. First, second, and third output power cost coefficients; , These represent traditional generating units. The startup and shutdown cost coefficients, Indicates traditional units The binary start / stop status; This represents the total number of new energy power stations in the new power system; Indicates new energy power station The cost coefficient of the disposal penalty; , They represent new energy power stations. The predicted output power and the actual output power.

[0045] In practical implementation, the operating constraints of the new power system can be expressed as follows: ; ; ; ; ; ; ; ; ; In the formula, For conventional units During the period contribution; For nodes During the period The load; For nodes During the period The load; For power transmission lines During the period The transmission power; , The units Lower and upper limits of output power; For new energy power stations During the period The predicted output power; , They are nodes and nodes During the period phase angle, To balance the nodes in the time period phase angle, For the line During the period The reactance; For the line Maximum transmission power capacity; binary unit During the period The running state variables; binary unit During the period The running status; , The units During the period Continuous power-on and power-off time, , The units Minimum startup time and minimum downtime constraints; , The units Upward and downward ramp output power limits.

[0046] The parameters required in the unit combination optimization model of the above-mentioned new power system can all be calculated based on the predicted curves of day-ahead renewable energy output and load, and obtained from the equipment parameters of each unit.

[0047] The steps for online identification of system inertia shortage using a new power system unit combination optimization model are as follows: S41. Obtain the day-ahead renewable energy output and load forecast curves, and obtain the day-ahead unit combination scheme through the unit combination optimization model.

[0048] S42. Filter the input data features in the day-ahead unit combination scheme to form an input data sequence and input it into the identification model.

[0049] The input data characteristics include: the total system inertia level, the system synchronous inertia level, the system virtual inertia level, and the total primary frequency regulation rate of the governor during each time period within the scheduling cycle, which can be expressed as: ; ; ; ; In the formula, , , , These represent the time periods. The system's synchronous inertia level, virtual inertia level, total inertia level, and total primary frequency regulation rate of the speed controller are all considered. Indicates the first The inertia constant of the synchronous generator unit Indicates the first Taiwan synchronous generator unit during the time period The binary start / stop status within. Indicates the first Taiwan synchronous generator unit during the time period Internal output power, Indicates the first A new energy unit during the period The virtual inertia constant within, Indicates the first A new energy unit during the period The actual output power inside, Indicates the first The speed governor of the synchronous generator unit during the time period The frequency modulation rate within.

[0050] S43. The identification model verifies the inertia level of the scheme by quickly predicting and judging key frequency characteristics.

[0051] The basis for determining whether the system inertia level is insufficient is whether the above-mentioned key frequency characteristic inequality holds. If it holds, it indicates that the system inertia level is insufficient and inertia resources need to be replenished to the system in a timely manner.

[0052] Compared with existing technologies, this embodiment provides a power system inertia shortage identification method based on frequency feature extraction. By constructing a system frequency swing equation that considers the virtual inertia response delay of new energy grid configurations, key frequency features for identifying system inertia level shortages are extracted. Then, a system inertia level shortage identification model is built using deep learning algorithms, and the model is trained offline based on historical unit combination data to improve the reliability and speed of online identification. Next, a new power system unit combination optimization model is constructed, and based on the day-ahead new energy output and load forecast curves, a day-ahead unit combination scheme is proposed. Finally, the proposed identification model is used to verify the inertia level of the scheme on an hourly time scale. By using the system's steady-state operating data to perform post-verification of the day-ahead unit combination scheme, the method can ensure the frequency security of the new power system operation.

[0053] Example 2 A specific embodiment of the present invention discloses a novel power system inertia level shortage identification system, comprising: The system frequency swing equation module is used to reflect the network-type virtual inertia response delay provided by new energy sources in the new power system, and to extract key frequency features that can accurately identify system inertia shortages. The inertia shortage identification module is used to build an inertia level shortage identification model for a new type of power system and train it offline to achieve online identification of system inertia level shortage. The day-ahead unit combination optimization module is used to provide day-ahead unit combination schemes based on the day-ahead renewable energy output and load forecast curves, and then use the identification model to verify the inertia level of the scheme.

[0054] The specific implementation process of this embodiment can be found in the above method embodiments, and will not be repeated here. Since this embodiment is based on the same principle as the above method embodiments, this system also has the corresponding technical effects of the above method embodiments.

[0055] Example 3 To verify the effectiveness and correctness of the solutions proposed in Examples 1 and 2, a specific embodiment 3 of the present invention was constructed using an improved North American NPCC-140 node system for verification. The system includes 140 nodes, 233 lines, 44 generator sets, and 4 new energy power stations, with the network structure as follows: Figure 2As shown in the figure, the system has a conventional turbine installed capacity of 18.05 GW and a wind power installed capacity of 8 GW, with a new energy penetration rate of 30.71%. The rated frequency is 60 Hz, with some conventional turbines providing frequency support and wind turbines participating in frequency response through VSG control technology. Based on historical data, typical daily wind power and load data were selected, and wind power output was appropriately increased. Referring to power system frequency safety regulations, the maximum frequency change rate limit was set at 0.5 Hz / s. Power disturbance was assumed to be 10% of the load in each time period.

[0056] First, K-means clustering was used to aggregate typical days from historical renewable energy and load data. An input sample set of 5600 data points was constructed through screening and feature selection. In the transient simulation, disturbances were randomly selected as 5% to 10% of the total system load level and distributed to each node according to the node load rate. Considering the adjustability of the virtual inertia control parameters, the VSG virtual inertia constant was randomly set to 4 to 12 s. The load surge was set to occur at 1 s, and the simulation duration was 15 s. Based on the constructed dataset, the recognition model was trained offline, with 180 training propagations performed at a batch size of 32. The distribution of the obtained prediction results is as follows: Figure 3 As shown, the prediction errors for the vast majority of key frequency change features are above 0, indicating that the security loss function constructed during the offline training process of the model is effective, making the prediction results of the identification model for key frequency change features more reliable.

[0057] The verification is performed using a sample from the test set. The selected day-ahead wind power output and load forecast curves are shown below. Figure 4 As shown in the figure. A basic day-ahead unit combination scheme was obtained through the establishment of a novel power system unit combination optimization model. Then, by identifying the model, the key frequency variation characteristics of the system were predicted, and the results are as follows. Figure 5 As shown, the basic day-ahead unit combination model was identified as having inertia shortages during certain time periods. The labeling results after identification are as follows. Figure 6 As shown.

[0058] The results of the embodiments show that the proposed novel power system inertia level shortage identification method can verify the inertia level of the scheme on an hourly time scale, and can ensure the frequency security of the operation of the novel power system.

[0059] Therefore, this invention adopts a power system inertia shortage identification method based on frequency feature extraction with the above-mentioned structure. It uses deep learning algorithms to build a model and achieves online identification of system inertia shortage after offline training based on historical unit combination data. This solves the problem that it is difficult to accurately identify system inertia level shortage caused by the access of new energy sources in the prior art, and provides a reference for subsequent inertia level supplementation planning work, which can ensure the frequency security of the operation of new power systems.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for identifying power system inertia shortage based on frequency feature extraction, characterized in that, Includes the following steps: S1. Construct a system frequency swing equation that considers the virtual inertia response delay of new energy grid-type systems, so as to extract key frequency features that can accurately identify the system inertia level shortage. S2. Using deep learning algorithms, a new identification model for power system inertia level shortage is constructed to achieve online identification of system inertia level shortage; S3. The identification model is trained offline using historical unit combination data to improve the reliability and speed of the model identification results; S4. Construct a new power system unit combination optimization model. Based on the day-ahead renewable energy output and load forecast curves, provide a day-ahead unit combination scheme, and then use the trained identification model to verify the inertia level of the scheme.

2. The power system inertia shortage identification method based on frequency feature extraction according to claim 1, characterized in that, In step S1, the system frequency swing equation considering the virtual inertia response delay of the new energy network is expressed as: ； In the formula, The first-order response delay for virtual inertia in new energy network structures; , These are the system's synchronous inertia energy and virtual inertia energy, respectively. , These are the primary frequency regulation power and the unbalanced power of the system, respectively. The damping coefficient of the system; The system's rated frequency; This refers to the frequency change of the system. This represents the time elapsed after the disturbance occurred.

3. The power system inertia shortage identification method based on frequency feature extraction according to claim 2, characterized in that, In step S1, the key frequency features extracted to identify the inertia level shortage of the identification system are: ； In the formula, The time when the disturbance occurs. This represents the maximum rate of frequency change that the power system can withstand. If any of the above key frequency characteristics are met, it indicates a shortage of system inertia, requiring timely replenishment of inertia resources.

4. The power system inertia shortage identification method based on frequency feature extraction according to claim 1, characterized in that, In step S2, a multi-layer neural network structure with convolutional computing capability and data feature dimensionality reduction capability is used to build the main network, which is then connected in series with a network layer with data shape compression capability and a network layer with multi-neuron connection capability, and finally connected in series with a functional module with key frequency feature judgment capability to form the recognition model.

5. The power system inertia shortage identification method based on frequency feature extraction according to claim 1, characterized in that, Step S3 includes the following sub-steps: S31. Obtain historical unit combination data, filter the input data types required for training, and form offline training dataset input data; S32. Based on historical unit combination data, set different disturbance powers in transient simulation, obtain the system frequency change curves corresponding to each operating moment in batches, extract the system frequency change curves, obtain the maximum frequency change rate of the system in batches, and form offline training dataset output data. S33. Combine the input and output data of the offline training dataset to form the offline training dataset; The offline training dataset is divided into a training set and a test set according to a certain ratio; S34. Set the optimizer, training batch, learning rate, number of training iterations and safety loss function during the training process. Based on the offline training test set performance of the recognition model, obtain the optimal inertia level shortage prediction and recognition model when the set stopping conditions are met.

6. The power system inertia shortage identification method based on frequency feature extraction according to claim 5, characterized in that, In step S34, the security loss function is expressed as: ； In the formula, This represents the loss function of the neural network in the recognition model. This indicates the recognition model in terms of network hyperparameters. The loss function used to predict the maximum rate of change of the system frequency is as follows.

7. The power system inertia shortage identification method based on frequency feature extraction according to claim 1, characterized in that, In step S4, the unit combination optimization model is as follows: under the constraints of the new power system operation, with the minimum operating cost as the objective function, the starting mode of each traditional unit, the output power of each traditional unit, and the actual output power of each new energy power station are obtained in each time period within the dispatch cycle; wherein, the objective function L is expressed as: ； In the formula, This represents the total cost of the output power of a synchronous generator unit. This represents the total cost of starting and stopping a synchronous generator unit. This represents the total cost of penalties for absorbing new energy power plants.

8. The power system inertia shortage identification method based on frequency feature extraction according to claim 7, characterized in that, Total cost of the output power of the synchronous generator unit Total start-up and shutdown costs of synchronous generator units Total cost of absorbing renewable energy power plants They are represented as follows: ； ； ； In the formula, Indicates the total number of time periods. This represents the total number of traditional generating units in the new power system; , , These represent traditional generating units. First, second, and third output power cost coefficients; , These represent traditional generating units. The startup and shutdown cost coefficients, Indicates traditional units The binary start / stop status; This represents the total number of new energy power stations in the new power system; Indicates new energy power station The cost coefficient of the disposal penalty; , They represent new energy power stations. The predicted output power and the actual output power.

9. The power system inertia shortage identification method based on frequency feature extraction according to claim 1, characterized in that, In step S4, the inertia shortage of the system is identified online through the following sub-steps: S41. Obtain the day-ahead renewable energy output and load forecast curves, and obtain the day-ahead unit combination scheme through the unit combination optimization model; S42. Filter the input data features in the day-ahead unit combination scheme to form an input data sequence and input it into the identification model; S43. The identification model verifies the inertia level of the scheme by quickly predicting and judging key frequency features.

10. The power system inertia shortage identification method based on frequency feature extraction according to claim 1, characterized in that, A system for implementing the method includes: The system frequency swing equation module is used to reflect the network-type virtual inertia response delay provided by new energy sources in the new power system, and to extract key frequency features that can accurately identify system inertia shortages. The inertia shortage identification module is used to build an inertia level shortage identification model for a new type of power system and train it offline to achieve online identification of system inertia level shortage. The day-ahead unit combination optimization module is used to provide a day-ahead unit combination scheme based on the day-ahead renewable energy output and load forecast curves, and then use the identification model to verify the inertia level of the scheme.

Images