A voltage sensitivity fitting method for a substation in a topological unknown state

Abstract

The application discloses a voltage sensitivity fitting method in a topological unknown state of a transformer area, and the fitting method comprises the following steps: a power-voltage fitting model is established, a generalized regression neural network is adopted, and the relationship between the power injection of the transformer area and the voltage is fitted; after the power-voltage fitting model is trained in step one, the relationship between the power change amount and the voltage change amount is obtained, and a voltage sensitivity fitting model is established; after the voltage sensitivity fitting model is trained, the voltage sensitivity obtained by fitting the real-time measurement data; the fitted voltage sensitivity is applied to online optimization, the adjustment amount of energy storage is solved according to the obtained voltage sensitivity, and the voltage is optimized. The fitting method can obtain the voltage sensitivity of controllable resources by using only part of real-time measurement in the absence of topology, and the voltage out-of-limit problem caused by the uncertainty and volatility of distributed photovoltaics can be optimized according to the sensitivity, and the voltage can be ensured not to be out of limit after optimization.

Description

Technical Field

[0001] This invention relates to the field of power distribution network optimization technology, specifically a voltage sensitivity fitting method for transformer substations in an unknown topology state. Background Technology

[0002] In 2020, the installed capacity of renewable energy sources such as photovoltaics and wind power in the power system increased significantly. However, due to the randomness and volatility of distributed power sources, their high proportion of decentralized access also brings many risks to the distribution network, such as voltage exceeding limits and increased fluctuation risks, significantly reducing power supply stability and having a significant adverse impact on the operational safety of low-voltage distribution areas. Using voltage sensitivity can quickly optimize the voltage of distribution areas and reduce voltage fluctuations. However, the solution for sensitivity is based on topology parameters, which are normally difficult to obtain in distribution areas.

[0003] Based on domestic and international research and applications, there are two main approaches to voltage optimization in transformer substations due to incomplete topology: fitting power flow based on neural networks to obtain a fitted power flow that replaces the actual physical power flow (fitted power flow optimization), and topology-aware optimization using sensing methods to obtain the substation's topology parameters and establish a power flow model. Fitted power flow optimization adapts well to substations with stable operating conditions and provides a solid foundation for voltage optimization. However, its effectiveness decreases under scenarios with significant power injection fluctuations, such as new photovoltaic installations, and it cannot guarantee that the substation's operating voltage will not exceed limits. Topology-aware optimization, on the other hand, uses a power flow model established by sensing topology parameters to adapt to changes in operating conditions, such as new photovoltaic installations, and can optimize voltage in scenarios with large power injection variations. However, when the actual topology changes, the topology parameters need to be re-sensed, and since topology changes in substations are frequent, the computational load for topology-aware optimization is too high. Furthermore, some real-time data for the substation is missing. Under these circumstances, neither of these two approaches can provide adequate support for substation voltage optimization. Therefore, this paper proposes a voltage sensitivity fitting method for substations with incomplete topology. Summary of the Invention

[0004] The purpose of this invention is to provide a voltage sensitivity fitting method for transformer substations in a topologically unknown state. By combining artificial intelligence methods with historical and real-time data of the substation operation, the voltage sensitivity of the substation is obtained through fitting, which is of great significance for ensuring the voltage quality of the substation. In the absence of topology, the voltage sensitivity of controllable resources is obtained by using only partial real-time measurements. Based on the sensitivity, the voltage limit exceeding problem caused by the uncertainty and volatility of distributed photovoltaics can be optimized, and the optimized voltage can be guaranteed not to exceed the limit.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] A voltage sensitivity fitting method for a topologically undefined plateau region, the fitting method comprising the following steps:

[0007] Step 1: Establish a power-voltage fitting model and use a generalized regression neural network to fit the relationship between the injected power and voltage in the transformer area.

[0008] Step 2: After training the power-voltage fitting model in Step 1, obtain the relationship between the corresponding power change and voltage change, and establish a voltage sensitivity fitting model.

[0009] Step 3: After training, the voltage sensitivity fitting model is used to fit the voltage sensitivity obtained from the real-time measurement data.

[0010] Step 4: Apply the fitted voltage sensitivity to online optimization. Solve for the energy storage adjustment amount based on the obtained voltage sensitivity to optimize the voltage.

[0011] Furthermore, the specific method for establishing the power-voltage fitting model is as follows:

[0012] (1) A power-voltage fitting model was established using a generalized regression neural network.

[0013] (2) Power-voltage fitting model: First, the GRNN network architecture is established, which consists of an input layer, a mode layer, a summation layer, and an output layer.

[0014] (3) Based on the constructed network, the power-voltage fitting model is trained using historical power-voltage data. The training objective is to minimize the root mean square error, with the root mean square error as the reference.

[0015]

[0016] In the formula: M is the number of samples in the test set, y test For the true value, y GRNN This is a voltage fitting model based on GRNN.

[0017] (4) After the model training is completed, the voltage of each node can be obtained by inputting power data, and the model can adapt to the changes in injected power within a certain range.

[0018] Furthermore, the voltage sensitivity fitting model is established using an SVM construction method as follows:

[0019] By simplifying the voltage sensitivity approximation, we obtain an approximate expression for the voltage sensitivity:

[0020]

[0021] In the formula: S U-P For active-voltage sensitivity, SU-Q This refers to reactive power-voltage sensitivity.

[0022] The voltage sensitivity of the transformer area under non-real-time observation conditions was obtained by using SVM fitting:

[0023] First, set the input and output of the voltage sensitivity fitting model: In the sensitivity fitting model, the input is the available real-time power information, and the output is the voltage sensitivity.

[0024] Then, the model is set to use an algorithm: the voltage sensitivity fitting model matches some real-time measurement information with historical operating states to obtain the voltage sensitivity information under that state, thus realizing real-time sensing of voltage sensitivity.

[0025] The final kernel function was set to be a linear kernel function.

[0026] Furthermore, the training of the voltage sensitivity fitting model based on the power-voltage fitting model is carried out as follows:

[0027] (1) Obtain the voltage sensitivity under different power conditions.

[0028] (2) Initialize the SVM using random initialization.

[0029] (3) Set the number of batch samples for one training session, randomly sample the same number of power data from the dataset and input them into the SVM, output the sensitivity of the fit, and compare it with the sensitivity in the dataset to optimize the model.

[0030] Furthermore, the method for applying the fitted voltage sensitivity to online optimization is as follows:

[0031] First, the voltage sensitivity of the transformer area is fitted. When applying it online, the trained voltage sensitivity fitting model is deployed to the transformer area, and the real-time node measurement data is then input into the fitting model.

[0032] Then, optimization is performed to obtain the voltage sensitivity. Based on the sensitivity, optimization commands are issued to the controllable resources in the transformer area to achieve voltage stability in the transformer area.

[0033] 6. Furthermore, the method of establishing a power-voltage fitting model using a generalized regression neural network includes:

[0034] (1) In power systems, due to the need to satisfy power flow constraints, there is an implicit functional relationship between node injected power and node voltage, as shown in the following equation:

[0035]

[0036] Where: G ij and B ijLet U be the real and imaginary parts of the line admittance between nodes i and j. i Let θ be the voltage magnitude at node i. ij P represents the voltage phase angle difference between nodes i and j. i and Q i These represent the injected active and reactive power at node i, respectively.

[0037] (2) After Taylor series expansion, the corrected equation is obtained as follows:

[0038]

[0039] Where J is the Jacobian matrix.

[0040] (3) In power systems, the changes in power and voltage can be directly solved when the topology parameters are known. When the transformer area topology is not available, potential power flow information can be mined from the data using a GRNN network, thereby establishing a mapping relationship from node injected power to node voltage. This voltage fitting model replaces the physical power flow model, providing a prerequisite for subsequently mining sensitivity information from it.

[0041] Furthermore, the input and output layers of the GRNN network architecture are structured as shown in the following equation:

[0042]

[0043] In the formula: X and Y are network input and output matrices, and k is the number of nodes in the distribution network excluding the slack node.

[0044] The input layer receives the injected active power P and reactive power Q from each node, with n = 2k neurons. In the pattern layer, each neuron corresponds to a different learning sample, with the number of neurons equal to the sample size m. The transfer function is shown in the following equation:

[0045]

[0046] In the formula, σ is the smoothness factor.

[0047] The summation layer consists of two different types of neurons, which perform summation and weighted summation on all neurons in the pattern layer, respectively:

[0048] The direct summation formula is:

[0049]

[0050] The weighted summation formula is:

[0051]

[0052] The output layer outputs the voltage of each node, and each neuron divides the output of the summation layer by the output of the summation layer.

[0053]

[0054] The beneficial effects of this invention are:

[0055] 1. The fitting method of this invention combines artificial intelligence with historical and real-time data of transformer substation operation to obtain the voltage sensitivity of the transformer substation, which is of great significance for ensuring the voltage quality of the transformer substation.

[0056] 2. The fitting method of this invention obtains the voltage sensitivity of controllable resources by using only some real-time measurements in the absence of topology. Based on the sensitivity, the voltage limit problem caused by the uncertainty and volatility of distributed photovoltaics can be optimized, and the optimized method can ensure that the voltage does not exceed the limit. Attached Figure Description

[0057] The invention will now be further described with reference to the accompanying drawings.

[0058] Figure 1 This is a schematic diagram of the topology of the transformer area in the example of this invention;

[0059] Figure 2 This is the voltage distribution diagram before optimization in this invention;

[0060] Figure 3 This is a thermal map of the actual sensitivity of the transformer area in this invention;

[0061] Figure 4 This is a heatmap of the fitting sensitivity of the transformer area in this invention;

[0062] Figure 5 This is the fitting sensitivity diagram under non-real-time observation conditions of the present invention;

[0063] Figure 6 This is the optimized voltage distribution diagram of the present invention. Detailed Implementation

[0064] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] A voltage sensitivity fitting method for topologically ambiguous mesas, such as Figure 1 The structure shown is optimized for non-real-time observation areas. In terms of data, the actual user load provided by the UCI machine learning library and the photovoltaic power output data provided by ELIA are used. The voltage data of each node is obtained through power flow calculation to simulate the actual operating state of the area.

[0066] This embodiment uses 96*90 sets of data from a non-real-time observation area, sampled every 15 minutes, for a total of 3 months, as the dataset. One day's data is randomly selected as an example to obtain the voltage before optimization. Figure 2 As shown.

[0067] The fitting method includes the following steps:

[0068] Step 1: First, based on the historical data of the transformer area's operation, establish a power-voltage fitting model. Use a generalized regression neural network to fit the relationship between the injected power and voltage of the transformer area. The specific method for establishing the power-voltage fitting model is as follows:

[0069] (1) A power-voltage fitting model was established using a generalized regression neural network.

[0070] In power systems, due to the need to satisfy power flow constraints, there is an implicit functional relationship between node injected power and node voltage, as shown in the following equation:

[0071]

[0072] Where: G ij and B ij U represents the real and imaginary parts of the line admittance between nodes i and j; i Let θ be the voltage magnitude at node i; ij P represents the voltage phase angle difference between nodes i and j; i and Q i These represent the injected active and reactive power at node i, respectively.

[0073] After Taylor series expansion, the corrected equation is obtained as follows:

[0074]

[0075] Where J is the Jacobian matrix.

[0076] It can be seen that in power systems, the changes in power and voltage can be directly solved when the topology parameters are known. However, when the transformer area topology is not available, other methods are needed to obtain power flow information. This invention uses a GRNN network to mine potential power flow information in the data, and then establishes a mapping relationship from node injected power to node voltage. This voltage fitting model replaces the physical power flow model, providing a prerequisite for subsequently mining sensitivity information from it.

[0077] (2) Power-voltage fitting model: First, the GRNN network architecture is established, which consists of an input layer, a pattern layer, a summation layer, and an output layer.

[0078] The network's input and output are shown in the following equations:

[0079]

[0080] In the formula: X and Y are network input and output matrices, and k is the number of nodes in the distribution network excluding the slack node.

[0081] The input layer receives the injected active power P and reactive power Q from each node, with n = 2k neurons. In the pattern layer, each neuron corresponds to a different learning sample, with the number of neurons equal to the sample size m. The transfer function is shown in the following equation:

[0082]

[0083] In the formula, σ is the smoothness factor.

[0084] The summation layer consists of two different types of neurons, which perform summation and weighted summation on all neurons in the pattern layer, respectively:

[0085] The direct summation formula is:

[0086]

[0087] The weighted summation formula is:

[0088]

[0089] The output layer outputs the voltage of each node, and each neuron divides the output of the summation layer by the output of the summation layer.

[0090]

[0091] (3) Based on the constructed network, the power-voltage fitting model is trained using historical power-voltage data. The training objective is to minimize the root mean square error, with the root mean square error as the reference.

[0092]

[0093] In the formula: M is the number of samples in the test set, y test For the true value, y GRNN This is a voltage fitting model based on GRNN.

[0094] (4) After the model training is completed, the voltage of each node can be obtained by inputting power data, and the model can adapt to the changes in injected power within a certain range.

[0095] Step 2: After training the power-voltage fitting model in Step 1, obtain the relationship between the corresponding power change and voltage change, and establish a voltage sensitivity fitting model.

[0096] By randomly selecting cross-sectional power data and generating several sets of power changes, these are input into the power-voltage fitting model to obtain voltage changes. The corresponding power and voltage changes are stored as the dataset for the voltage sensitivity fitting model. Based on real-time operating conditions, the actual voltage sensitivity is obtained. Taking active voltage sensitivity as an example, see attached figure. Figure 3 As shown.

[0097] A voltage sensitivity fitting model is constructed using SVM, and the specific construction method is as follows:

[0098] By simplifying the voltage sensitivity approximation, we can obtain an approximate expression for the voltage sensitivity:

[0099]

[0100] In the formula: S U-P For active-voltage sensitivity, S U-Q This refers to reactive power-voltage sensitivity.

[0101] Voltage sensitivity is related to topology parameters and injected power. When real-time measurement data is complete, the least squares method is used to fit and solve the voltage sensitivity to obtain the active and reactive voltage sensitivity of the transformer area. Since it is difficult to obtain complete real-time measurement data in the transformer area, the voltage sensitivity of the transformer area under non-real-time observation conditions is obtained by using SVM fitting.

[0102] First, set the input and output of the voltage sensitivity fitting model: In the sensitivity fitting model, the input is the available real-time power information, and the output is the voltage sensitivity.

[0103] Then, the model is set to use an algorithm: the voltage sensitivity fitting model matches some real-time measurement information with historical operating states to obtain the voltage sensitivity information under that state, thus realizing real-time sensing of voltage sensitivity.

[0104] This invention uses SVM to fit voltage sensitivity. SVM is a machine learning method based on statistical learning theory and the principle of structural risk minimization. It has many unique advantages in solving small sample, nonlinear and high-dimensional pattern recognition problems, and largely overcomes problems such as "curse of dimensionality" and "overlearning". It has a good effect on fitting voltage sensitivity.

[0105] The final kernel function was set to be a linear kernel function.

[0106] Step 3: Train the voltage sensitivity fitting model based on the datasets of power and voltage changes. Input the measurement data to obtain the voltage sensitivity of each node. After training, the voltage sensitivity fitting model will fit the voltage sensitivity obtained from real-time measurements as follows: Figure 4 As shown

[0107] The fitted voltage sensitivity has a small error compared to the actual sensitivity, and it can satisfy the actual voltage sensitivity distribution law, allowing for voltage optimization. The comparison between the fitted voltage sensitivity and the actual sensitivity under non-real-time observation conditions is as follows: Figure 5 As shown, high-precision fitting of voltage sensitivity can be achieved even when only partial real-time data is available.

[0108] The voltage sensitivity fitting model is trained based on the power-voltage fitting model, and the specific steps are as follows:

[0109] (1) Obtain voltage sensitivity under different power conditions

[0110] A set of cross-sectional data is randomly obtained from historical data. Random fluctuations are added to the power data, and this data is input into a power-voltage fitting model to obtain the voltage at each node. The fitted voltage is compared with the actual cross-sectional voltage to obtain the corresponding voltage fluctuation. The relationship between ΔP, ΔQ, and ΔU under P and Q states at this cross-section is obtained, and the voltage zero sensitivity S is calculated. U-P S U-Q ; Select several sets of historical data and perform the same operation to obtain the voltage sensitivity under different cross sections, and store it in the dataset;

[0111] (2) Initialize the SVM using random initialization;

[0112] (3) Set the number of batch samples for one training session, randomly sample the same number of power data from the dataset and input them into the SVM, output the sensitivity of the fit, and compare it with the sensitivity in the dataset to optimize the model.

[0113] Step 4: In online application, input the real-time measurement data into the trained voltage sensitivity fitting model to obtain the voltage sensitivity. Based on the obtained voltage sensitivity, calculate the energy storage adjustment amount (energy storage within the transformer area) to optimize the voltage. The optimized voltage is as follows: Figure 6 As shown, the method for applying the fitted voltage sensitivity to online optimization is as follows:

[0114] First, the voltage sensitivity of the transformer area is fitted. When applying it online, the trained voltage sensitivity fitting model is deployed to the transformer area, and the real-time node measurement data is input into the fitting model.

[0115] Then, optimization is performed to obtain the voltage sensitivity. Based on the sensitivity, optimization commands are issued to the controllable resources in the transformer area to achieve voltage stability in the transformer area.

[0116] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0117] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention.

Claims

1. A method for fitting voltage sensitivity in a transformer area with an unknown topology, characterized in that, The fitting method includes the following steps: Step 1: Establish a power-voltage fitting model and use a generalized regression neural network to fit the relationship between the injected power and voltage in the transformer area; Step 2: After training the power-voltage fitting model in Step 1, obtain the relationship between the corresponding power change and voltage change, and establish a voltage sensitivity fitting model; Step 3: After training, the voltage sensitivity fitting model is used to fit the voltage sensitivity obtained from the real-time measurement data; Step 4: The fitted voltage sensitivity is applied to online optimization. Based on the obtained voltage sensitivity, the adjustment amount of energy storage is solved to optimize the voltage. The specific method for establishing the power-voltage fitting model is as follows: (1) A power-voltage fitting model was established using a generalized regression neural network; (2) Establish the GRNN network architecture, which consists of an input layer, a pattern layer, a summation layer, and an output layer; (3) Based on the constructed network architecture, the power-voltage fitting model is trained using historical power-voltage data. The training objective is to minimize the root mean square error. In the formula: The number of samples in the test set. For the true value, This is a voltage fitting model based on GRNN; (4) After the model training is completed, the voltage of each node can be obtained by inputting power data, and it can adapt to the changes in injected power within a certain range of amplitude. The voltage sensitivity fitting model is established using the SVM construction method as follows: By simplifying the voltage sensitivity approximation, we obtain an approximate expression for the voltage sensitivity: In the formula: For active-voltage sensitivity, Reactive power-voltage sensitivity; The voltage sensitivity of the transformer area under non-real-time observation conditions was obtained by using SVM fitting: First, set the input and output of the voltage sensitivity fitting model: In the sensitivity fitting model, the input is the available real-time power information, and the output is the voltage sensitivity; Then, the model is set to use an algorithm: the voltage sensitivity fitting model matches some real-time measurement information with historical operating states to obtain the voltage sensitivity information under that state, thus realizing real-time sensing of voltage sensitivity. The final kernel function was set to be a linear kernel function.

2. The voltage sensitivity fitting method for a transformer area in a topologically unknown state according to claim 1, characterized in that, The voltage sensitivity fitting model is trained based on the power-voltage fitting model, and the specific steps are as follows: (1) Obtain the voltage sensitivity under different power conditions; (2) Initialize the SVM using random initialization; (3) Set the number of batch samples for one training session, randomly sample the same number of power data from the dataset and input them into the SVM, output the sensitivity of the fit, and compare it with the sensitivity in the dataset to optimize the model.

3. The voltage sensitivity fitting method for a transformer area in a topologically unknown state according to claim 2, characterized in that, The method for applying the fitted voltage sensitivity to online optimization is as follows: First, the voltage sensitivity of the transformer area is fitted. When applying it online, the trained voltage sensitivity fitting model is deployed to the transformer area, and the real-time node measurement data is input into the fitting model. Then, optimization is performed to obtain the voltage sensitivity. Based on the sensitivity, optimization commands are issued to the controllable resources in the transformer area to achieve voltage stability in the transformer area.

4. The voltage sensitivity fitting method for a transformer area in an unknown topology as described in claim 1, characterized in that, The method of establishing a power-voltage fitting model using a generalized regression neural network includes: (1) In power systems, due to the need to satisfy power flow constraints, there is an implicit functional relationship between node injected power and node voltage, as shown in the following equation: In the formula: and For nodes 、 The real and imaginary parts of the line admittance; For nodes The voltage amplitude; For nodes , Voltage phase angle difference between them; and They are nodes The injected active and reactive power; (2) After Taylor series expansion, the corrected equation is obtained as follows: in It is a Jacobian matrix; (3) In the power system, the changes in power and voltage can be directly solved when the topology parameters are known. When the transformer area topology cannot be obtained, the potential power flow information in the data is mined by using the GRNN network, and then the mapping relationship from the injected power to the node voltage is established. The power-voltage fitting model replaces the physical power flow model, providing a prerequisite for the subsequent mining of sensitivity information.

5. The voltage sensitivity fitting method for a transformer area in an unknown topology state according to claim 4, characterized in that, The input and output layers of the GRNN network architecture are structured as follows: In the formula: , For network input and output matrices, This represents the number of nodes in the distribution network excluding the slack node. The input layer contains the injected active power of each node. and reactive power Number of neurons In the pattern layer, each neuron corresponds to a different learning sample, and the number of neurons is equal to the sample size. The transfer function is shown in the following equation: In the formula, It is the smoothing factor; The summation layer consists of two different types of neurons, which perform summation and weighted summation on all neurons in the pattern layer, respectively: The direct summation formula is: The weighted summation formula is: The output layer outputs the voltage of each node, and each neuron divides the output of the summation layer by the output of the summation layer. 。

Images