A transient power angle stability control decision method and system considering stability judgment confidence

Abstract

The present invention proposes a transient power angle stability control decision method and system taking into account the confidence of stability judgment, the method includes: constructing a set of fault scenarios, combining the operation mode with the set of fault scenarios for simulation and then performing stability judgment; intercepting the first time series data from the time series data after judgment as input, and the stability judgment result as output, combining the input and output to obtain sample data; intercepting the sub-sample data corresponding to each stability judgment network from the sample data according to the input length of each stability judgment network, and using the sub-sample data to train the stability judgment network; acquiring and accumulating real-time measurement data, calling the network to perform stability judgment when the length meets a certain network input requirement, and judging whether to implement emergency control according to the confidence, until the current network stability judgment result meets the confidence threshold or all networks are called. Based on this method, a transient power angle stability control decision system is also proposed. The present invention constructs multiple stability judgment networks with different input data window lengths to achieve both fast and accurate stability judgment.

Description

Technical Field

[0001] The invention belongs to the technical field of power angle stability control of electric power systems, and in particular relates to a transient power angle stability control decision method and system taking into account stability judgment confidence. Background Art

[0002] The safety and stability performance of the power system refers to the ability of the system to recover to its original state or transition to a new stable state after being disturbed, which is mainly guaranteed by the safety defense system. Among them, power angle stability refers to the ability of the synchronous generators in the system to maintain synchronous operation after the system is disturbed, which is an important part of the safety and stability of the power system. With the development of various renewable energy power generation and ultra-high voltage direct current transmission technology, the power system is gradually transitioning to a complex AC / DC hybrid power grid. On the one hand, the increasingly complex power system leads to an increased probability of transient stability-related failures and complex characteristics, and increased difficulty in stability judgment; on the other hand, it also leads to faster development of system instability. Once the system shows an unstable tendency, corresponding stabilization measures need to be taken as soon as possible to maintain the stability of the system. Therefore, how to quickly and accurately determine transient stability and take corresponding stability control measures is the focus of power system safety and stability analysis and control.

[0003] The transient stability judgment methods widely used in power systems mainly include the curve judgment method based on time domain simulation and the direct method based on the system dynamic model. The time domain simulation method obtains stability assessment by recording transient process data and using a computer to solve the system transient differential equation. It has high accuracy but takes a long time, especially for complex power systems, there will be a "dimensionality curse" problem. The direct method does not solve the system transient differential equation, but directly obtains the transient stability of the system based on a series of numerical calculation characteristics in the transient process, such as the transient energy method. Compared with the time domain simulation method, the direct method does not need to solve high-dimensional state differential equations, which significantly reduces the solution time, but it has a limited scope of application due to its reliance on strong assumptions. It is difficult to use it for complex power system analysis and control. In recent years, thanks to the progress of communication and information processing technology, the power grid control center has gradually acquired the ability to quickly obtain more comprehensive state information of the power system, and data-driven transient stability judgment methods have become the development direction, such as artificial neural networks, support vector machines, decision trees, etc. Among them, machine learning algorithms based on artificial neural networks have gradually become an important development direction. This method first constructs a stability judgment network with time series data as input, and then uses a machine learning algorithm to train the stability judgment network to automatically extract stability features and judge stability. However, this method usually takes time series data of fixed length as input, so there is a specific inherent delay in stability judgment. When a serious fault occurs in the system, it cannot use rapidly developing features to quickly judge stability, and cannot meet the requirements of emergency control as quickly as possible. Therefore, the existing data-driven stability judgment method relies on time series data sequences for stability judgment, and it needs to accumulate to a set time length before stability mapping can be performed. At the same time, in order to ensure the accuracy of stability judgment, it usually takes a long accumulation time, which leads to emergency control being too late and too costly in the scenario of rapidly developing faults. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a transient power angle stability control decision method and system taking into account the stability judgment confidence level, adopts multiple stability judgment networks to replace the multi-round stability judgment method of a single stability judgment network, and organically combines the multiple stability judgment networks through the stability judgment confidence level to realize the transient power angle stability control decision taking into account the stability judgment confidence level while taking into account the rapidity and accuracy of the stability judgment.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A transient power angle stability control decision method taking into account stability judgment confidence comprises the following steps:

[0007] Construct a fault scenario set for power system power angle stability judgment, perform dynamic time domain simulation on the simulation scenario set of each operation mode combined with the fault scenario set, perform transient stability judgment on the first time series data obtained by the simulation to complete stability judgment and mark;

[0008] Extract the second time series data from the time when the fault is removed to the end of the simulation from the first time series data after the marking is completed, and combine the second time series data with the marking value to obtain sample data; determine the input data length of each stability judgment network according to the total length of the sample data and the required stability judgment rounds, extract the corresponding length from the sample data to form sub-sample data corresponding to each stability judgment network, and use the sub-sample data to train each network to determine the parameters of each stability judgment network;

[0009] Acquire and accumulate real-time measurement data. When the length of the measurement data meets the input requirement of a stability judgment network, call the network to perform stability judgment. Determine whether to implement emergency control measures based on the stability judgment confidence level until the stability judgment result of the current stability judgment network meets the confidence threshold or all networks have been called, thereby completing online stability judgment and emergency control.

[0010] A transient power angle stability control decision system taking into account stability judgment confidence comprises a preprocessing module, a stability judgment data processing module and an online stability judgment module;

[0011] The preprocessing module is used to construct a fault scenario set for power system power angle stability judgment, perform dynamic time domain simulation on a simulation scenario set composed of each operation mode and the fault scenario set, perform transient stability judgment on the first time series data obtained by the simulation, complete stability judgment and mark;

[0012] The stability judgment data processing module is used to intercept the second time series data from the time when the fault is removed to the end of the simulation from the first time series data after the marking is completed, and combine the second time series data with the marking value to obtain sample data; determine the input data length of each stability judgment network according to the total length of the sample data and the required stability judgment rounds, intercept the corresponding length from the sample data to form sub-sample data corresponding to each stability judgment network, and use the sub-sample data to train each network to determine the parameters of each stability judgment network;

[0013] The online stability judgment module is used to obtain and accumulate real-time measurement data. When the length of the measurement data meets the input requirement of a certain stability judgment network, the network is called to perform stability judgment. Whether to implement emergency control measures is determined based on the stability judgment confidence level until the stability judgment result of the current stability judgment network meets the confidence threshold or all networks have been called, thereby completing online stability judgment and emergency control.

[0014] The effects provided in the content of the invention are only the effects of the embodiments, not all the effects of the invention. One of the above technical solutions has the following advantages or beneficial effects:

[0015] The present invention proposes a transient power angle stability control decision method and system taking into account the confidence of stability judgment, belonging to the technical field of power angle stability control of electric power system. The method comprises the following steps: constructing a fault scenario set for power angle stability judgment of electric power system, performing dynamic time domain simulation on a simulation scenario set composed of each operation mode and the fault scenario set, performing transient stability judgment on the first time series data obtained by simulation to complete stability judgment and mark; intercepting the second time series data from the fault removal to the end time of simulation from the first time series data after marking, and combining the second time series data with the mark value to obtain sample data; determining the input data length of each stability judgment network according to the total length of the sample data and the required stability judgment rounds, intercepting the corresponding length from the sample data to form sub-sample data corresponding to each stability judgment network, using the sub-sample data to train each network to determine the parameters of each stability judgment network; acquiring and accumulating real-time measurement data, calling the network to perform stability judgment when the measurement data length meets the input requirement of a certain stability judgment network, judging whether to implement emergency control measures according to the stability judgment confidence, until the stability judgment result of the current stability judgment network meets the confidence threshold or all networks have been called, thereby completing online stability judgment and emergency control. Based on a transient power angle stability control decision method taking into account the confidence of stability judgment, a transient power angle stability control decision system taking into account the confidence of stability judgment is also proposed. The present invention adopts multiple improved one-dimensional convolutional neural networks with different lengths of time series data input as stability judgment networks, and sequentially calls the corresponding stability judgment networks for stability judgment according to the length of the currently acquired time series data, and executes the stability control scheme when the confidence meets the set threshold. Compared with the stability judgment based on a single neural network based on fixed-length time series data, the present invention can dynamically adjust the export time of the stability judgment result according to the real-time characteristics of the system, taking into account the stability judgment delay and the stability judgment accuracy, and realizing rapid stability judgment and stabilization control based on a small amount of measurement data in the case of serious faults to avoid excessive control costs, and accurate stability judgment based on more measurement data for non-serious faults to avoid false operation, supporting the construction of a new type of power system stability control system.

[0016] The present invention includes two stages: construction of multiple stability judgment networks (offline stage) and multi-round stability judgment control (online application). The offline stage mainly generates a data set covering application scenarios and constructs multiple stability judgment networks with different input data window lengths. The online application stage sequentially forms input data adapted to different networks based on measurement information and performs cyclic stability judgment. When the stability judgment confidence meets the requirements, emergency control is initiated to realize the online application of multiple stability judgment networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A general flow chart of the prior art neural network stability determination;

[0018] Figure 2 This is an overall flow chart of a transient power angle stability control decision method taking into account stability judgment confidence proposed in Embodiment 1 of the present invention;

[0019] Figure 3 A flowchart of the offline part of a transient power angle stability control decision method taking into account the stability judgment confidence proposed in Example 1 of the present invention;

[0020] Figure 4 A flowchart of the online part of a transient power angle stability control decision method taking into account the stability judgment confidence proposed in Example 1 of the present invention;

[0021] Figure 5 A schematic diagram of a data sample assembly process for network construction proposed in Embodiment 1 of the present invention;

[0022] Figure 6 This is a structural diagram of the improved one-dimensional convolutional neural network proposed in Example 1 of the present invention;

[0023] Figure 7 This is a schematic diagram of the application example results proposed in Example 1 of the present invention;

[0024] Figure 8 This is a schematic diagram of a transient power angle stability control decision system taking into account stability judgment confidence proposed in Example 2 of the present invention. DETAILED DESCRIPTION

[0025] In order to clearly illustrate the technical features of the present solution, the present invention is described in detail below through specific implementation methods and in conjunction with the accompanying drawings. The disclosure below provides many different embodiments or examples for realizing different structures of the present invention. In order to simplify the disclosure of the present invention, the components and settings of specific examples are described below. In addition, the present invention may repeat reference numbers and / or letters in different examples. This repetition is for the purpose of simplification and clarity, and does not itself indicate the relationship between the various embodiments and / or settings discussed. It should be noted that the components illustrated in the accompanying drawings are not necessarily drawn to scale. The present invention omits the description of known components and processing techniques and processes to avoid unnecessary limitations on the present invention.

[0026] Example 1

[0027] Embodiment 1 of the present invention proposes a transient power angle stability control decision method taking into account the stability judgment confidence. The implementation basis of the present invention is a stability judgment method based on an artificial neural network. Figure 1The general flow chart of the neural network stability judgment of the prior art; first, generate sample data for network training and verification according to the neural network application scenario; second, use correlation analysis, principal component analysis, clustering and other methods to extract key feature quantities that affect the output of the mapping network to achieve network simplification; third, construct an input feature set based on the key feature quantities, organize the input data set for network construction according to the feature set structure and mark the output target value of each data sample; fourth, divide the data set containing the input features and the corresponding target output into a training set and a verification and test set, determine the parameters of each node of the neural network based on the training set using training methods such as maximum gradient descent, and use the verification and test sets to evaluate the network performance; fifth, construct the network input in the form of a feature set based on real-time measurement information, and use the neural network mapping to obtain the output to achieve the neural network application.

[0028] Transient stability judgment of power system is a typical binary classification problem, and its output results only include two state judgments: stable or unstable. Input data mainly includes two types: one is single-dimensional information, which may be the value of system characteristic information at a certain time section or the comprehensive value over a period of time; the other is multi-dimensional information, which generally sorts the selected key feature quantities according to time to form time series data, so that the neural network input data is a two-dimensional matrix arranged according to the type of feature quantity and time. Usually, the single-dimensional input data form is mainly used for simpler artificial neural networks, which contains limited data features and may affect the accuracy of stability judgment when used for complex problems; the multi-dimensional input data form contains the time domain variation law of each feature quantity and can reflect richer information. It is usually used in complex artificial neural networks (such as convolutional neural networks, long short-term memory networks, etc.) for complex problem feature extraction.

[0029] As the scale of power systems increases, the types of components increase, and the characteristics become more complex, the stability judgment network using multi-dimensional input data has become the mainstream. The stability judgment network using multi-dimensional input data needs to accumulate a certain length of input data to better capture the system characteristics. Reasonable selection of the time window for accumulating time series data is an important part of the design of the stability judgment network. A longer time series data window can ensure the accuracy of the stability judgment, but it will bring a larger emergency control delay. Excessive control delay may further lead to a significant increase in the cost of emergency control, and even cross the point of no return and fail to restore the power angle stability of the system; a shorter time series data window can quickly give the stability judgment result and perform emergency control, but insufficient input data can easily lead to a decrease in the accuracy of the stability judgment, resulting in missed judgment (instability is judged as stability) and causing system instability or misjudgment (stability is judged as instability) causing additional machine shedding and load shedding.

[0030] Therefore, the present invention adopts multiple stability judgment networks to replace the multi-round stability judgment method of a single stability judgment network, and organically combines multiple stability judgment networks through stability judgment confidence, so as to realize transient power angle stability control decision taking into account the stability judgment confidence while taking into account the rapidity and accuracy of stability judgment. The present invention includes two stages: multi-stability judgment network construction (offline stage) and multi-round stability judgment control (online application): the offline stage mainly generates a data set covering the application scenario, and constructs multiple stability judgment networks with different input data window lengths; the online application stage sequentially forms input data adapted to different networks based on measurement information and performs cyclic stability judgment, and starts emergency control when the stability judgment confidence meets the requirements, so as to realize the online application of multiple stability judgment networks.

[0031] Figure 2 This is an overall flow chart of a transient power angle stability control decision method taking into account stability judgment confidence proposed in Embodiment 1 of the present invention;

[0032] In step S200, a fault scenario set for power system power angle stability judgment is constructed, dynamic time domain simulation is performed on the simulation scenario set composed of each operation mode and the fault scenario set, transient stability judgment is performed on the first time series data obtained by the simulation to complete stability judgment and mark it.

[0033] In step S210, the second time series data from the time when the fault is removed to the end of the simulation is intercepted from the first time series data after the marking is completed, and the second time series data is combined with the marking value to obtain sample data; the input data length of each stability judgment network is determined according to the total length of the sample data and the required stability judgment rounds, and the corresponding length is intercepted from the sample data to form sub-sample data corresponding to each stability judgment network, and each network is trained using the sub-sample data to determine the parameters of each stability judgment network.

[0034] In step S220, real-time measurement data is acquired and accumulated. When the length of the measurement data meets the input requirement of a certain stability judgment network, the network is called to perform stability judgment. Whether to implement emergency control measures is determined based on the stability judgment confidence level until the stability judgment result of the current stability judgment network meets the confidence threshold or all networks have been called, thereby completing online stability judgment and emergency control.

[0035] Figure 3 A flowchart of the offline part of a transient power angle stability control decision method taking into account the stability judgment confidence proposed in Example 1 of the present invention;

[0036] In step 1, a fault scenario set for power system power angle stability judgment is constructed, and dynamic time domain simulation is performed on the simulation scenario set composed of each operation mode and the fault scenario set. The transient stability judgment is performed on the first time series data obtained by the simulation to complete the stability judgment and mark it.

[0037] The fault scenario set for power system power angle stability judgment includes:

[0038] The grid topology is set up considering the grid structure that may have transient power angle stability problems. Based on any grid topology, possible load levels are set, and the matching generator startup mode and output level form an operation scenario set; that is: investigate the historical operation data of the target power system to form a grid topology set under typical modes and typical load curves under various topologies; according to the load curve and the power supply and energy storage configuration of the power system, an economic dispatch method is used to determine the generator startup status and the corresponding power generation output, energy storage charging and discharging power under each load level to form an operation scenario set.

[0039] For the load level, generator startup status and corresponding power generation output, energy storage charging and discharging power status under any topology, statistics are collected on generator, load, transformer, line and bus power and line load rate; based on the statistical results, the top 10% of line power and load rate, the top 1% of bus power throughput, the top 1% of node load power, the top 1% of generator output and the top 1% of transformer power are selected as key components.

[0040] Faults related to power angle stability are applied one by one to each key component in the operation scenario set to form a fault scenario set. The faults related to power angle stability specifically include: three-phase short circuit and disconnection of key lines, three-phase short circuit and disconnection of key buses, a sudden increase in load of 100%, fault-free tripping of generators, disconnection after three-phase short circuit on the high-voltage side of transformers, fault-free disconnection of all renewable energy sites, single-pole and double-pole locking of DC transmission lines, and power drop.

[0041] The specific process of dynamic time domain simulation of the simulation scenario set composed of each operation mode and fault scenario set includes: combining each operation scenario with any fault in the corresponding fault scenario set to obtain all simulation scenario sets to be simulated; using commercial electromechanical transient software (such as BPA, PSS / E, etc.) to simulate the transient process of the power system for all simulation scenarios, recording the power angle, speed, mechanical input power and electromagnetic output power of all generators during the simulation process, the frequency, voltage amplitude and voltage phase angle of all nodes, the active and reactive power flow of all lines and other time series data, storing the recorded results as the first time series data, and naming them according to the operation scenario and fault scenario.

[0042] The stability judgment for the first time series data is completed and marked as a mark value using a mechanism-based stability judgment method, and the specific process includes: using a mechanism-based stability judgment method (such as maximum power angle difference, extended equal area development (EEAC)) and other methods to process the simulation data to complete stability judgment and marking. The calculation method based on EEAC is given in this application as an example. That is: read the generator speed and power angle data in each simulation result record, use a clustering method to group them, and further equate them to a single-machine infinite system; based on the equivalent single-machine infinite system, use the EEAC theory to calculate the acceleration and deceleration area difference (when the system is unstable) or the potential deceleration area (when the system is stable) to form a power angle stability margin; according to whether the power angle stability margin is greater than zero, mark the simulation data sample output Y as 1 (stable) or 0 (unstable).

[0043] In step 2, the second time series data from the time when the fault is removed to the end of the simulation is intercepted from the first time series data after the marking is completed, and the second time series data is combined with the marking value to obtain sample data; the input data length of each stability judgment network is determined according to the total length of the sample data and the required stability judgment rounds, and the corresponding length is intercepted from the sample data to form sub-sample data corresponding to each stability judgment network.

[0044] The sub-time series data are arranged in rows in terms of time and in columns in terms of categories to form sample data for stability judgment; Figure 5 This is a schematic diagram of the process of assembling a data sample for network construction proposed in Embodiment 1 of the present invention; namely:

[0045]

[0046] Among them, A i (i=1,…n), B i (i=1,…m), C i (i=1,…s) is the time series data vector of various key characteristic quantities, such as the power angle, speed, power, etc. of each generator; a it , b it (t=1, ...T) are the first key feature quantities a i and the second key feature quantity B i The time series takes values ​​at time t, where T is the total number of moments;

[0047] Scan all sample data, and do not return point data index L when statistics are unstable i , and take the minimum value L=min(L i ) is used as the maximum length of the input data of the stability judgment network.

[0048] According to the number of rounds of stability determination required for multiple rounds of stability determination, the input data length L1, L2...L of each stability determination network is determined. N The method is to take half or divide equally.

[0049] The formula for taking half in sequence is:

[0050]

[0051] The averaging method is:

[0052]

[0053] In step 3, a multi-stability judgment network is constructed. Figure 6 This is a structural diagram of the improved one-dimensional convolutional neural network proposed in Example 1 of the present invention;

[0054] In this application, the multi-stability judgment network adopts an improved one-dimensional convolutional neural network model; the improved one-dimensional convolutional neural network model includes 3 convolutional layers, 3 pooling layers, 2 fully connected layers and 1 output layer;

[0055] The convolution layer uses 5 5×5 convolution filters and the maximum pooling function. The data structure of each convolution layer is the same as the input data length L. i Correspondingly; the output layer uses the softmax function for binary classification, and selects the ReLU function as the neuron activation function.

[0056] Set network training hyperparameters such as learning rate, maximum number of iterations, loss function, and error propagation method;

[0057] The corresponding stability judgment network is trained using the sub-sample data to obtain the trained stability judgment network set. The sub-data set for construction is divided into training set, validation set and test set according to 7:2:1. The neural network training is carried out using neural network construction software such as matlab and pytorch. The accuracy of the training set and the accuracy of the validation set are monitored as the training rounds change. When the progress of the stability judgment network on the two data sets enters a stable state, it is considered that the i-th stability judgment network has been completed.

[0058] Determine whether all stable networks have been constructed. If not, continue to build the next stable network. If yes, end the offline training phase and output N stable networks.

[0059] The online stage is mainly to determine whether the system has power angle instability based on real-time measurement information when power angle disturbance events such as power system failures occur, and to initiate pre-designed emergency control measures when the system is unstable. Figure 4 A flowchart of the online part of a transient power angle stability control decision method taking into account the stability judgment confidence proposed in Example 1 of the present invention;

[0060] In step 4, set the input data length L of the stability judgment network during multiple rounds of stability judgment i and the confidence threshold C ri, and set the stable network index i to 1 to indicate the stable network number to be used.

[0061] In step 5, real-time measurement data is acquired and accumulated. When the length of the measurement data meets the input requirements of a stability judgment network, the network is called to perform stability judgment. Specifically, the electrical quantity data after fault removal is accumulated according to the stability judgment network input data format until the accumulated data length meets the current stability judgment network input data length requirement, the input data matrix is ​​constructed according to the stability judgment network input data format, and the i-th stability judgment network is called to map the stability judgment output; when judging whether the output data meets the preset confidence threshold, the current stability judgment confidence C is calculated based on the softmax output. i :

[0062] C i =max(p i1 ,p i2 )·|p i1 -p i2 |

[0063] In the formula, p i1 ,p i2 are the probabilities of stable and unstable results output by the softmax function respectively;

[0064] Then the current confidence level C i With confidence threshold C ri Compare and determine whether the preset confidence threshold is met. If so, execute step 7; if not, execute step 6.

[0065] In step 6, the current stable network index is increased by 1 to determine whether the current stable network index exceeds the total number of networks. If so, go to step 7. If not, go back to step 5 and use the next stable network to determine stability.

[0066] In step 7, stability control is performed according to the current stability judgment result. If the judgment result is that the system power angle is stable, no control is performed and the emergency control process is reset. If the judgment result is that the system power angle is unstable, the emergency control strategy table is searched and the corresponding machine shedding and load shedding control measures are started.

[0067] Embodiment 1 of the present invention proposes a transient power angle stability control decision method taking into account the stability judgment confidence level, which adopts multiple stability judgment networks to replace the multi-round stability judgment method of a single stability judgment network, and organically combines the multiple stability judgment networks through the stability judgment confidence level to achieve a transient power angle stability control decision taking into account the stability judgment confidence level while taking into account the rapidity and accuracy of the stability judgment.

[0068] In order to explain in detail the implementation process of a transient power angle stability control decision method taking into account the stability judgment confidence proposed in Example 1 of the present invention, the transient power angle stability control process of the IEEE39 node standard example system is taken as an example to demonstrate the effectiveness of the method proposed in the present invention. Figure 5 The improved one-dimensional convolutional neural network structure shown in the figure convolves the input data according to the physical meaning. The application process is as follows:

[0069] Step 1: Assume that the IEEE30 system only uses the initial topology, that is, the topology change is not considered, and the load of each node fluctuates within 90% to 130% of the typical load level. Five groups of load levels are randomly selected within the fluctuation range, and the corresponding generator output is determined by economic dispatch, forming a total of 300 initial operation modes; the line branch between node 25 and node 26 is selected as the key element, and a three-phase ground short circuit fault is set and applied to the head end of the branch. The fault time is 1s after the start of the simulation, the fault duration is 0.15 to 0.24s, the interval is 0.01s, and there are a total of 10 different fault removal times. When the fault is removed, the circuit breakers on both sides of the fault line are actuated, and the entire line is removed.

[0070] Combining the aforementioned initial operation mode and fault setting, PSS / E was used to perform electromechanical transient simulation. The simulation results recorded included the generator power angle and speed of 10 generators, 30 bus voltage amplitudes and phase angles, and active and reactive bidirectional power flow data of 46 lines, with a total of 2,890 simulation sample data.

[0071] Based on the EEAC theory, the simulation data curves are equalized, the stability margin is calculated, and marked as stable or unstable.

[0072] Step 2: Network construction is built using sample data sets. Statistical analysis of all samples shows that the fastest instability time is 100ms after the fault is removed. Considering a certain stability judgment delay, three stability judgment neural networks are set up, and the input data lengths are 5ms, 40ms and 80ms respectively.

[0073] Step 3: The network model adopts an improved one-dimensional convolutional neural network model. The one-dimensional convolutional neural network is a special convolutional neural network, which is generally composed of several convolutional layers, pooling layers, ReLU layers (activation function layers) and a group of fully connected layers. The specific structure is shown in Figure 5 .

[0074] The original sample set is cut in the time series direction of the data, and matched with the input layer of the short time series network and the input layer of the long time series network respectively, to generate independent sample sets. After all samples are shuffled in the independent sample sets of each network, the first 2500 samples are taken as the training set, and the last 390 samples are taken as the test set. The training / test sample ratio is about 6:1. According to the training steps, the short time series network and the long time series network are trained separately.

[0075] Figure 7 This is a schematic diagram of the application example results proposed in Example 1 of the present invention; when applied online, the time series data is accumulated and the stability judgment network is called one by one to judge stability. The examples are as follows:

[0076] Operation mode 1, fault duration 0.23s, stability confidence threshold set to 0.8.

[0077] The short-time network judgment result is unstable, with a confidence level of 0.9955, which meets the threshold output and directly outputs control without using the subsequent network to judge stability. At this time, the emergency control judgment delay is about 5ms, which can significantly reduce the cost of subsequent stable control measures.

[0078] Operation mode 2, fault duration 0.24s, stability confidence threshold set to 0.8.

[0079] When the data accumulation time reaches 5ms, the judgment result is stable, but the confidence is 0.6740, which does not meet the threshold, so the time series data continues to be accumulated for the next round of stability judgment; when the data accumulation time reaches 40ms, the judgment result is stable, and the confidence is 0.9953, which meets the threshold requirement, and the stability control ends.

[0080] Operation mode 3, fault duration 0.19s, stability confidence threshold set to 0.8.

[0081] When the data accumulation time reaches 5ms, the judgment result is stable, but the confidence level is 0.5368, which does not meet the threshold, so the time series data continues to be accumulated for the next round of stable judgment; when the data accumulation time reaches 40ms, the judgment result is unstable, and the confidence level is 0.8578, which meets the threshold requirement, and emergency control measures are initiated.

[0082] Example 2

[0083] Based on the transient power angle stability control decision method taking into account the stability judgment confidence proposed in Example 1 of the present invention, Example 2 of the present invention further proposes a transient power angle stability control decision system taking into account the stability judgment confidence. Figure 8 A schematic diagram of a transient power angle stability control decision system taking into account stability judgment confidence proposed in Embodiment 2 of the present invention, comprising a preprocessing module, a stability judgment data processing module and an online stability judgment module;

[0084] The preprocessing module is used to construct a fault scenario set for power system power angle stability judgment, perform dynamic time domain simulation on the simulation scenario set of each operation mode and fault scenario set combination, perform transient stability judgment on the first time series data obtained by simulation, complete stability judgment and mark;

[0085] The stability judgment data processing module is used to intercept the second time series data from the time when the fault is removed to the end of the simulation from the first time series data after the marking is completed, and combine the second time series data with the marking value to obtain sample data; determine the input data length of each stability judgment network according to the total length of the sample data and the required stability judgment rounds, intercept the corresponding length from the sample data to form sub-sample data corresponding to each stability judgment network, and use the sub-sample data to train each network to determine the parameters of each stability judgment network;

[0086] The online stability judgment module is used to obtain and accumulate real-time measurement data. When the length of the measurement data meets the input requirement of a certain stability judgment network, the network is called to perform stability judgment. Whether to implement emergency control measures is determined based on the stability judgment confidence level until the stability judgment result of the current stability judgment network meets the confidence threshold or all networks have been called, thereby completing online stability judgment and emergency control.

[0087] In the preprocessing module, the process of constructing a fault scenario set for power angle stability judgment of the power system includes: setting a power grid topology structure considering the power grid structure where transient power angle stability problems may occur, setting possible load levels based on any power grid topology structure, and matching generator startup methods and output levels to form an operation scenario set; applying faults related to power angle stability to each key component in the operation scenario set one by one to form a fault scenario set, and the faults related to power angle stability specifically include: three-phase short circuit and disconnection of key lines, three-phase short circuit and disconnection of key buses, sudden increase of load by 100%, fault-free tripping of generators, disconnection after three-phase short circuit on the high-voltage side of transformers, fault-free disconnection of all renewable energy sites, single-pole and double-pole locking of DC transmission lines, and power drop.

[0088] The specific process of performing transient stability judgment on the first time series data obtained by simulation, completing stability judgment and marking includes:

[0089] Combine each operation scenario with any fault in the corresponding fault scenario set to obtain a simulation scenario set to be simulated; simulate the transient process of the power system for all simulation scenarios and then output the first time series data;

[0090] A mechanism-based stability judgment method is used to complete stability judgment for the first time series data and mark it as a mark value; specifically: read the generator speed and power angle data in each simulation result record, use a clustering method to group them, and further equate them to a single-machine infinite system; based on the equivalent single-machine infinite system, a mechanism-based stability judgment method is used to calculate the acceleration and deceleration area difference or the potential deceleration area to form a power angle stability margin; according to whether the power angle stability margin is greater than zero, the first time series data output is marked as stable or unstable.

[0091] In the stability judgment data processing module, the process of intercepting the corresponding length from the sample data to form sub-sample data corresponding to each stability judgment network includes:

[0092] The sample data is arranged in rows in time and in columns in categories to form sub-sample data for stability judgment; that is:

[0093]

[0094] Among them, A i (i=1,…n), B i (i=1,…m), C i (i=1, ...s) is the time series data vector of various key feature quantities; a it , b it (t=1, ...T) are the first key feature quantities a i and the second key feature quantity B i The time series takes values ​​at time t, where T is the total number of moments;

[0095] Scan all sample data, and do not return point data index L when statistics are unstable i , and take the minimum value L=min(L i ) is used as the maximum length of the input data of the stability judgment network.

[0096] Input the sample data into the L i The rows are combined with the stable labels to form the sub-sample data for network training.

[0097] According to the number of rounds of stability determination required for multiple rounds of stability determination, the input data length L1, L2...L of each stability determination network is determined. N The method is to take half or divide equally;

[0098] The formula for taking half in sequence is:

[0099]

[0100] The averaging method is:

[0101]

[0102] The stability judgment network adopts an improved one-dimensional convolutional neural network model; the improved one-dimensional convolutional neural network model includes 3 convolutional layers, 3 pooling layers, 2 fully connected layers and 1 output layer;

[0103] The convolution layer uses 5 5×5 convolution filters and a maximum pooling function. The data structure of each convolution layer is consistent with the input data length L i Correspondingly; the output layer uses the softmax function for binary classification, and selects the ReLU function as the neuron activation function.

[0104] In the online stability judgment module, when using the trained stability judgment network set for stability judgment, the input data length L of the stability judgment network during multiple rounds of stability judgment is set i and the confidence threshold C ri And the stable network index i is set to 1 to indicate the stable network number to be used currently.

[0105] The specific process of acquiring and accumulating real-time measurement data and calling a network to perform stability judgment when the length of the measurement data meets the input requirement of a stability judgment network includes: acquiring and accumulating real-time measurement data, and calling the network to judge when the length of the measurement data meets the input requirement of the i-th stability judgment network to obtain the stability judgment result and confidence. The stability judgment confidence Ci is calculated based on the softmax output:

[0106] C i =max(p i1 ,p i 2)·|p i1 -p i2 |; In the formula, p i1 ,p i2 are the probabilities of stable and unstable results output by the softmax function respectively;

[0107] Then the current confidence level C i With confidence threshold C ri Compare and determine whether the preset confidence threshold is met.

[0108] The current confidence level C i With confidence threshold C ri The specific process of comparing and judging whether the preset confidence threshold is met is as follows: when the confidence C given by the i-th stability judgment network is i When the preset confidence threshold is met, the corresponding stability control measures are implemented according to the stability judgment result to complete the online stability control process. If the requirements are not met, the data continues to be accumulated and the i+1th stability judgment network is called for stability judgment and control until the stability judgment result of the i+1th stability judgment network meets the confidence threshold or all networks have been called.

[0109] Embodiment 2 of the present invention further proposes a transient power angle stability control decision system taking into account the confidence of stability judgment, adopts multiple improved one-dimensional convolutional neural networks with time series data inputs of different lengths as stability judgment networks, and calls the corresponding stability judgment networks in sequence according to the length of the currently acquired time series data to perform stability judgment, and executes the stability control scheme when the confidence meets the set threshold. Compared with the stability judgment based on a single neural network based on fixed-length time series data, the present invention can dynamically adjust the export time of the stability judgment result according to the real-time characteristics of the system, taking into account the stability judgment delay and stability judgment accuracy, and realizing rapid stability judgment and stabilization control based on a small amount of measurement data in the event of a serious fault to avoid excessive control costs, and accurate stability judgment based on more measurement data in the event of a non-serious fault to avoid false operation, thereby supporting the construction of a new type of power system stability control system.

[0110] The description of the relevant parts of a transient power angle stabilization control decision system taking into account the stability judgment confidence level provided in Example 2 of the present application can be found in the detailed description of the corresponding parts of a transient power angle stabilization control decision method taking into account the stability judgment confidence level provided in Example 1 of the present application, and will not be repeated here.

[0111] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the term "include", "comprise" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or equipment that includes a series of elements are inherent to the elements. In the absence of more restrictions, the elements limited by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or equipment that includes the elements. In addition, the above-mentioned technical solution provided in the embodiment of the present application is consistent with the corresponding technical solution in the prior art in principle, and the part is not described in detail, so as not to repeat too much.

[0112] Although the above describes the specific implementation of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. For those skilled in the art, other different forms of modifications or deformations can be made on the basis of the above description. It is not necessary and impossible to list all the implementation methods here. On the basis of the technical solution of the present invention, various modifications or deformations that can be made by those skilled in the art without creative work are still within the scope of protection of the present invention.

Claims

1. A transient power angle stability control decision method taking into account the stability judgment confidence, characterized in that: The following steps are involved: Construct a fault scenario set for power system power angle stability judgment, perform dynamic time domain simulation on the simulation scenario set of each operation mode and the combination of the fault scenario set, perform transient stability judgment on the first time series data obtained by simulation, complete stability judgment and mark; wherein, the specific process of performing transient stability judgment on the first time series data obtained by simulation, complete stability judgment and mark includes: combining each operation scenario with any fault in the corresponding fault scenario set to obtain a simulation scenario set to be simulated; simulating the transient process of the power system for all simulation scenarios and then outputting the first time series data; using a mechanism-based stability judgment method to complete stability judgment for the first time series data and mark it as a mark value; specifically: reading the generator speed and power angle data in each simulation result record, using a clustering method to group them, and further equating them to a single-machine infinite system; based on the equivalent single-machine infinite system, using a mechanism-based stability judgment method to calculate the acceleration and deceleration area difference or potential deceleration area to form a power angle stability margin; according to whether the power angle stability margin is greater than zero, mark the first time series data output as stable or unstable; The second time series data from the time when the fault is removed to the end of the simulation is intercepted from the first time series data after the marking is completed, and the second time series data is combined with the marking value to obtain sample data; the input data length of each stability judgment network is determined according to the total length of the sample data and the required stability judgment rounds, and the corresponding length is intercepted from the sample data to form sub-sample data corresponding to each stability judgment network, and the sub-sample data is used to train each network to determine the parameters of each stability judgment network; the process of intercepting the corresponding length from the sample data to form the sub-sample data corresponding to each stability judgment network includes: The sample data are arranged in rows in time and in columns in categories to form sub-sample data for determining stability; that is: Among them, A i (i=1,…n), B i (i=1,…m), C i (i=1, ...s) is the time series data vector of various key feature quantities; a it , b it (t=1, ...T) are the first key feature quantities A i and the second key feature quantity B i The time series takes values ​​at time t, where T is the total number of moments; Scan all sample data, and do not return point data index L when statistics are unstable i , and take the minimum value L=min(L i ) as the maximum length of the input data of the stability judgment network; Input the sample data into the L i The rows are combined with the stable labels to form the sub-sample data for network training; Determine the input data length L1, L2, ... L of each stability judgment network according to the number of stability judgment rounds required for determining multiple rounds of stability judgment. N The method is to take half or divide equally; The formula for taking half in sequence is: L N =L, The averaging method is: L N =L, Acquire and accumulate real-time measurement data. When the length of the measurement data meets the input requirement of a stability judgment network, call the network to perform stability judgment. Determine whether to implement emergency control measures based on the stability judgment confidence level until the stability judgment result of the current stability judgment network meets the confidence threshold or all networks have been called, thereby completing online stability judgment and emergency control.

2. The transient power angle stability control decision method taking into account the stability judgment confidence according to claim 1 is characterized in that: The process of constructing a fault scenario set for determining power system power angle stability includes: Considering the grid structure where transient power angle stability problems may occur, the grid topology is set, and based on any grid topology, possible load levels are set, as well as matching generator startup modes and output levels to form an operation scenario set; Faults related to power angle stability are applied one by one to each key component in the operation scenario set to form a fault scenario set. The faults related to power angle stability specifically include: three-phase short circuit and disconnection of key lines, three-phase short circuit and disconnection of key buses, a sudden increase in load of 100%, fault-free tripping of generators, disconnection after three-phase short circuit on the high-voltage side of transformers, fault-free disconnection of all renewable energy sites, single-pole and double-pole locking of DC transmission lines, and power drop.

3. The transient power angle stability control decision method taking into account the stability judgment confidence according to claim 1 is characterized in that: The stability judgment network adopts an improved one-dimensional convolutional neural network model; the improved one-dimensional convolutional neural network model includes 3 convolutional layers, 3 pooling layers, 2 fully connected layers and 1 output layer; The convolution layer uses 5 5×5 convolution filters and a maximum pooling function. The data structure of each convolution layer is consistent with the input data length L i Correspondingly; the output layer uses the softmax function for binary classification, and selects the ReLU function as the neuron activation function.

4. The transient power angle stability control decision method taking into account the stability judgment confidence level according to claim 3 is characterized in that: When using the trained stability judgment network set for stability judgment, set the input data length L of the stability judgment network during multiple rounds of stability judgment i and the confidence threshold C ri , and set the stable network index i to 1 to indicate the stable network number to be used.

5. The transient power angle stability control decision method taking into account the stability judgment confidence according to claim 4 is characterized in that: The specific process of acquiring and accumulating real-time measurement data and calling the network to perform stability judgment when the length of the measurement data meets the input requirement of a certain stability judgment network includes: acquiring and accumulating real-time measurement data, calling the network to judge when the length of the measurement data meets the input requirement of the i-th stability judgment network to obtain a stability judgment result and confidence, and the stability judgment confidence C i Calculation based on softmax output: C i =max(p i1 ,p i2 )·|p i1 -p i2 | In the formula, p i1 ,p i2 are the probabilities of stable and unstable results output by the softmax function respectively; Then the current confidence level C i With confidence threshold C ri Compare and determine whether the preset confidence threshold is met.

6. The transient power angle stability control decision method taking into account the stability judgment confidence level according to claim 5 is characterized in that: The current stability confidence C i With confidence threshold C ri The specific process of comparing and judging whether the preset confidence threshold is met is as follows: when the confidence C given by the i-th stability judgment network is i When the preset confidence threshold is met, the corresponding stability control measures are implemented according to the stability judgment result to complete the online stability control process. If the requirements are not met, the data continues to be accumulated and the i+1th stability judgment network is called for stability judgment and control until the stability judgment result of the i+1th stability judgment network meets the confidence threshold or all networks have been called.

7. A transient power angle stabilization control decision system taking into account the stability judgment confidence level, used to execute a transient power angle stabilization control decision method taking into account the stability judgment confidence level as claimed in any one of claims 1 to 6, characterized in that: It includes a preprocessing module, a stability judgment data processing module and an online stability judgment module; The preprocessing module is used to construct a fault scenario set for power system power angle stability judgment, perform dynamic time domain simulation on a simulation scenario set composed of each operation mode and the fault scenario set, perform transient stability judgment on the first time series data obtained by the simulation, complete stability judgment and mark; The stability judgment data processing module is used to intercept the second time series data from the time when the fault is removed to the end of the simulation from the first time series data after the marking is completed, and combine the second time series data with the marking value to obtain sample data; determine the input data length of each stability judgment network according to the total length of the sample data and the required stability judgment rounds, intercept the corresponding length from the sample data to form sub-sample data corresponding to each stability judgment network, and use the sub-sample data to train each network to determine the parameters of each stability judgment network; The online stability judgment module is used to obtain and accumulate real-time measurement data. When the length of the measurement data meets the input requirement of a certain stability judgment network, the network is called to perform stability judgment. Whether to implement emergency control measures is determined based on the stability judgment confidence level until the stability judgment result of the current stability judgment network meets the confidence threshold or all networks have been called, thereby completing online stability judgment and emergency control.

Images