Multi-point Error Parameter Location Method for Power Flow Calculation Model Based on Random Climbing

Abstract

The present invention discloses a method for locating multi-point error parameters of a power flow calculation model based on random hill climbing, which can accurately calibrate the dynamic parameters of a generator, thereby providing a reliable reference quantity for real-time fault detection and optimal control of the generator to ensure the normal and safe operation of the power grid. It is very simple to operate and can be completed in a short time.

Description

Technical Field

[0001] The present invention relates to the field of safe and stable operation and optimal dispatching of power grids, and particularly to a method for locating multi-point error parameters of a power flow calculation model based on stochastic ramp. Background Art

[0002] The safe operation of a power grid depends on accurate model parameters for power system analysis, simulation, and fault diagnosis. Therefore, reliable model parameters are the basis for power grid safe operation analysis.

[0003] Among the existing parameter identification methods, most are for identifying a single parameter in the power grid. When there are many incorrect parameters in the power grid model, although these methods can identify the incorrect parameters one by one, the operation is inevitably cumbersome and difficult to complete in a short time. Developing a method for locating multi-point error parameters is of great significance for power grid dispatching operation. Summary of the Invention

[0004] The present invention aims to overcome the deficiencies of the above-mentioned existing generator parameter calibration methods and provides a method for locating multi-point error parameters of a power flow calculation model based on stochastic ramp.

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

[0006] A method for locating multi-point error parameters of a power flow calculation model based on stochastic ramp, comprising the following steps:

[0007] Step 1: Obtain the actual power flow operation data of the power grid at a certain time section, including the active power and reactive power output by generators, the active power and reactive power of loads, the active power and reactive power flowing through each line and transformer, and the voltage amplitude and phase angle of each node;

[0008] Step 2: Calculate the power flow simulation data of the model of the power grid at this time section; in the simulation model of the power grid, through simulation calculation, obtain the power flow data corresponding to generators and loads in the power grid, the power flow flowing through each line and transformer, and the voltage and phase angle values of all nodes;

[0009] Step 3: Correspond the power flow operation data of the power grid obtained in Step 1 with the active power and reactive power of each device and the voltage amplitude and phase angle values of each node in the power flow calculation of the power grid model obtained in Step 2 one by one, and take the difference to obtain the absolute value of the power flow data error ΔX of all devices, and obtain the error data corresponding to all devices, where X is the active power P, reactive power Q, voltage amplitude U, or phase angle θ;

[0010] Step 4: Based on the above error data, screen out the area G with larger errors. Since the change of error parameters will cause the change of power flow in the power grid, first determine the values of ΔPmax, ΔQmax, ΔUmax, and Δθmax. The method for determining the error area G is as follows:

[0011] Determine a very small number e, where e = 0.01ΔXmax;

[0012] Judge whether a certain node or line is within the error area:

[0013] If X < e, this node or line is not within the error area;

[0014] If X ≥ e, this node or line is within the error area. When P, Q, U, and θ all meet the requirements, conduct a global search within the power grid, screen out the error areas, obtain multiple error areas, and number these areas as G1…Gk respectively;

[0015] Regard the generator as a PV node, the load as a PQ node, and the transformer adopts a τ-type equivalent model. Under normal circumstances, its excitation branch can be ignored. At this time, the transformer can be regarded as having nodes at both ends and a line in the middle. At this time, the error area becomes composed of connected nodes and lines. Draw the topology diagrams of all error areas;

[0016] Step 5: Randomly locate multiple power equipment within area G as the positioning equipment, and find all the equipment connected to the positioning equipment as the connected equipment;

[0017] Step 6: Through Step 3, obtain the power flow data errors ΔP and ΔQ of the positioning equipment and the connected equipment described in Step 5. Compare the power flow error of the positioning equipment with that of the connected equipment to judge whether the error of the connected equipment is larger than that of the positioning equipment;

[0018] Step 7: If there is no connected equipment with an error larger than that of the positioning equipment, determine this positioning equipment as the error equipment;

[0019] If there is a connected equipment with an error larger than that of the positioning equipment, select the direction with the largest increase in error value for searching, which is called climbing; locate the connected equipment with the largest error value as the new positioning equipment, then the original positioning equipment becomes the new connected equipment, continue to search for the connected equipment with the positioning equipment as the center, and go back to Step 5;

[0020] Step 8: Obtain all the error equipment within area G, and then train through the BP neural network to judge the error parameters of the error equipment and be able to locate the error parameters.

[0021] Furthermore, in the above Step 8, the specific method for identifying error parameters using the BP neural network is as follows:

[0022] Taking the line loss, the phase angle difference at both ends of the device, the change in the power flow through the device, and the change in the voltage amplitude at both ends of the device as the inputs of the neural network, accurately identify the error parameters of the device. The steps are as follows:

[0023] S1: Modify different parameters of the device through multiple experiments to generate errors, and collect the line loss, phase angle difference, power flow, and voltage amplitude changes caused by different errors;

[0024] S2: Normalize the change amounts obtained through step S1;

[0025] S3: Extract training samples and test samples from the characteristic quantities obtained through step S2;

[0026] S4: Construct a BP neural network;

[0027] S5: Train the BP neural network until a satisfactory accuracy is achieved;

[0028] S6: Test the BP neural network.

[0029] Furthermore: The training sample data in step S3 are respectively: the normalized data of the change amounts of line loss, phase angle difference, power flow, and voltage amplitude.

[0030] Furthermore: The data normalization method is as follows:

[0031] Let the power reference value in the power grid be SB and the voltage reference value be UB.

[0032]

[0033] Furthermore: The correspondence table between the training samples and the error parameter types is that if the code is 1000, the error parameter type is resistance error; if the code is 0100, the error parameter type is inductance error; if the code is 0010, the error parameter type is susceptance error; if the code is 0001, the error parameter type is turn ratio error. If there are multiple parameter errors, in the code, the corresponding positions are 1 and the rest are 0.

[0034] Furthermore: The BP neural network is a feedforward BP neural network with one hidden layer. Among them, the input layer has 5 neurons, which respectively correspond to the change amount data in the training samples, the output layer has 4 neurons, which respectively correspond to the error parameter types, and the number of neurons in the hidden layer is set to 8.

[0035] Furthermore: Fuzzify the neural network output data, set the data greater than 0.5 to 1, and the rest to 0.

[0036] Further: The neural network is trained according to the sample data using the trainlam function. The number of training times is 1000, and the learning rate is set to 0.005.

[0037] Further: In step 4:

[0038] Randomly locate several power equipment. The method is as follows:

[0039] Number all the nodes in all error regions G from 1 to n. According to the different numbers of nodes in each error region, divide the region Gi into mi small regions on average. Randomly select one node in each small region, then randomly select nodes from the entire error region G. Let the numbers of these nodes be:

[0040] a = {a1, a2,... aM};

[0041] Let the number of nodes in region Gi be si, mi = [0.1si], rounded down;

[0042] Select the nodes numbered with the elements in the array a and start the positioning search. The method of starting the positioning search from the node is as follows: First, find all the lines connected to the node, including transformers. Compare the power errors of these lines, select the line with the largest error as the positioning device, and conduct the next search for this positioning device. First, find the connected devices of the positioning device and conduct error comparison. If there is a connected device with an error larger than that of the positioning device, select the connected device with the largest error as the next positioning device. Finally, all the positioning searches will converge to some fixed positioning devices, and these devices are confirmed as error devices.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] The present invention can accurately calibrate the dynamic parameters of the generator, thereby providing a reliable reference quantity for the real-time fault detection and optimal control of the generator to ensure the normal and safe operation of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is the flow chart of the present invention;

[0046] Figure 2 is the error search flow chart in the embodiment provided by the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0047] Please refer to Figure 1 and Figure 2 , Figure 1 is the flow chart of the present invention; Figure 2 is the error search flow chart in the embodiment provided by the present invention.

[0048] To achieve the above object of the invention, the following technical solutions are adopted:

[0049] A method for locating multi-point error parameters of a power flow calculation model based on random hill climbing, comprising the following steps:

[0050] Step 1: Obtain the actual power flow operation data of the power grid at a certain time section, including the active power and reactive power output by generators, the active power and reactive power of loads, the active power and reactive power flowing through each line and transformer, and the voltage amplitude and phase angle of each node;

[0051] Step 2: Calculate the power flow simulation data of the model of the power grid at this time section; in the simulation model of the power grid, through simulation calculation, obtain the power flow data corresponding to generators and loads in the power grid, the power flow flowing through each line and transformer, and the voltage and phase angle values of all nodes;

[0052] Step 3: Correspond one by one the power grid operation power flow data obtained in Step 1 with the active power and reactive power of each device and the voltage amplitude and phase angle values of each node calculated by the power grid model in Step 2, and take the difference to obtain the absolute value of the power flow data error ΔX of all devices, and obtain the error data corresponding to all devices, where X is the active power P, reactive power Q, voltage amplitude U or phase angle θ;

[0053] Step 4: On the basis of the above error data, screen out the area G with larger errors. Since the change of error parameters will cause the change of power flow in the power grid, first determine the values of ΔPmax, ΔQmax, ΔUmax, and Δθmax. The method for determining the error area G is as follows:

[0054] Determine a very small number e, e = 0.01ΔXmax;

[0055] Judge whether a certain node or line is in the error area:

[0056] If X < e, this node or line is not in the error area;

[0057] If X ≥ e, this node or line is in the error area. When P, Q, U, and θ all meet the conditions, perform a global search in the power grid, screen out the error areas, obtain multiple error areas, and number these areas as G1…Gk respectively;

[0058] Regard the generator as a PV node, the load as a PQ node, and the transformer adopts a τ-type equivalent model. Under normal circumstances, its excitation branch can be ignored. At this time, the transformer can be regarded as having nodes at both ends and a line in the middle. At this time, the error area becomes composed of nodes and lines connected. Draw the topology diagrams of all error areas.

[0059] In the fourth step:

[0060] Randomly locate several power devices. The method is as follows:

[0061] Number all the nodes in all error regions G from 1 to n. According to the different numbers of nodes in each error region, divide the region Gi into mi small regions on average. Randomly select one node in each small region, then M nodes are randomly selected from the entire error region G. Let the numbers of these nodes be:

[0062] a = {a1, a2, … aM};

[0063] Let the number of nodes in region Gi be si, mi = [0.1si], rounded down;

[0064] Select the nodes numbered with the elements in the array a and start the positioning search. The method of starting the positioning search from the nodes is as follows: First, find out all the lines connected to the node, including transformers. Compare the power errors of these lines, select the line with the largest error as the positioning device, and conduct the next search for this positioning device. First, find out the connected devices of the positioning device and conduct error comparison. If there is a connected device with an error larger than that of the positioning device, select the connected device with the largest error as the next positioning device. Finally, all the positioning searches will converge to some fixed positioning devices, and these devices are confirmed as error devices.

[0065] Step Five: Randomly locate multiple power devices in region G as positioning devices, and find out all the devices connected to the positioning devices as connected devices;

[0066] Step Six: Through Step Three, the power flow data errors ΔP and ΔQ of the positioning devices and connected devices described in Step Five can be obtained. Compare the power flow errors of the positioning devices with those of the connected devices to determine whether the errors of the connected devices are larger than those of the positioning devices;

[0067] Step Seven: If there is no connected device with an error larger than that of the positioning device, determine the positioning device as an error device;

[0068] If there is a connected device with an error larger than that of the positioning device, select the direction with the largest increase in error value for search, which is called climbing; locate the connected device with the largest error value as the new positioning device, then the original positioning device serves as the new connected device, continue to search for connected devices with the positioning device as the center, and go to Step Five;

[0069] Step Eight: Obtain all the error devices in region G, and then train through a BP neural network to judge the error parameters of the error devices and be able to locate the error parameters.

[0070] The specific method for identifying the error parameters using a BP neural network is as follows:

[0071] Taking the line power loss, the phase angle difference at both ends of the device, the change in the power flow through the device, and the change in the voltage amplitude at both ends of the device as the inputs of the neural network, the error parameters of the device are accurately identified as follows:

[0072] S1: Modify different parameters of the device through multiple experiments to generate errors, and collect the line power loss, phase angle difference, power flow, and voltage amplitude changes caused by different errors;

[0073] S2: Normalize the change amounts obtained in step S1;

[0074] S3: Extract training samples and test samples from the feature quantities obtained in step S2;

[0075] S4: Construct a BP neural network;

[0076] S5: Train the BP neural network until a satisfactory accuracy is achieved;

[0077] S6: Test the BP neural network.

[0078] The training sample data in step S3 are respectively: the normalized data of the change amounts of the line power loss, phase angle difference, power flow, and voltage amplitude. The data normalization method is as follows:

[0079] Let the power reference value in the power grid be SB and the voltage reference value be UB.

[0080]

[0081]

[0082] The corresponding table of the training samples and the error parameter types is as follows: If the code is 1000, the error parameter type is resistance error; if the code is 0100, the error parameter type is inductance error; if the code is 0010, the error parameter type is susceptance error; if the code is 0001, the error parameter type is transformation ratio error. If there are multiple parameter errors, in the code, the corresponding positions are 1 and the rest are 0.

[0083] The described BP neural network is a feedforward BP neural network with one hidden layer. Among them, the input layer has 5 neurons, which respectively correspond to the change amount data in the training samples, and the output layer has 4 neurons, which respectively correspond to the error parameter types. The number of neurons in the hidden layer is set to 8.

[0084] Fuzzify the neural network output data, and set the data greater than 0.5 to 1 and the rest to 0.

[0085] Training the neural network based on the sample data is carried out using the trainlam function, with the number of training times being 1000 times and the learning rate set to 0.005.

[0086] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0087] The present invention can accurately calibrate the dynamic parameters of the generator, thereby providing a reliable reference quantity for the real-time fault detection and optimal control of the generator, ensuring the normal and safe operation of the power grid, and being very simple to operate and can be completed in a short time.

Claims

1. A method for locating multi - point error parameters of a power flow calculation model based on stochastic hill - climbing, characterized in that, It includes the following steps: Step 1: Obtain the actual power flow operation data at a certain time section of the power grid, including the active power and reactive power output by generators, the active power and reactive power of loads, the active power and reactive power flowing through each line and transformer, as well as the voltage amplitude and phase angle of each node; Step 2: Calculate the power flow simulation data of the power grid model at this time section; in the simulation model of this power grid, through simulation calculation, obtain the power flow data corresponding to generators and loads in the power grid, the power flow flowing through each line and transformer, as well as the voltage and phase angle values of all nodes; Step 3: One-to-one correspond the power flow operation data of the power grid obtained in Step 1 with the active power and reactive power of each device and the voltage amplitude and phase angle values of each node obtained from the power flow calculation of the power grid model in Step 2, and take the difference to obtain the absolute value of the power flow data error ΔX of all devices, obtaining the error data corresponding to all devices, where X is the active power P, reactive power Q, voltage amplitude U or phase angle θ; Step 4: Based on the above error data, screen out the regions G with larger errors. Since the change of error parameters will cause the change of power flow in the power grid, first determine the values of ΔPmax, ΔQmax, ΔUmax, and Δθmax. The method for determining the error region G is as follows: Determine a very small number e, e = 0.01ΔXmax; Judge whether a certain node or line is in the error region: If X < e, this node or line is not in the error region; If X ≥ e, this node or line is in the error region. When P, Q, U, and θ all meet the conditions, conduct a global search in the power grid, screen out the error regions, obtain multiple error regions, and number these regions as G1…Gk respectively; Regard the generator as a PV node, the load as a PQ node, and the transformer adopts a τ-type equivalent model. Under normal circumstances, its excitation branch can be ignored. At this time, the transformer can be regarded as having nodes at both ends and a line in the middle. At this time, the error region becomes composed of connected nodes and lines, and draw the topological diagrams of all error regions; Step 5: Randomly locate multiple power devices in region G as the located devices, and find all the devices connected to the located devices as the connected devices; Step 6: Through Step 3, the power flow data errors ΔP and ΔQ of the located devices and connected devices described in Step 5 can be obtained. Compare the power flow error of the located devices with the power flow error of the connected devices to judge whether the error of the connected devices is larger than the error of the located devices; Step 7: If there is no connected device with an error larger than that of the located device, determine this located device as the error device; If there is a connected device with an error larger than that of the located device, select the direction with the largest increase in error value for search, which is called climbing; locate the connected device with the largest error value as the new located device, then the original located device becomes the new connected device, continue to search for connected devices with the located device as the center, and go to Step 5; Step 8: Obtain all the error devices in region G, and then train through a BP neural network to judge the error parameters of the error devices and be able to locate the error parameters.

2. The method for locating multi - point error parameters of a power flow calculation model based on stochastic hill - climbing according to claim 1, characterized in that, In the eighth step described above, the specific method for identifying the error parameters using a BP neural network is as follows: Taking the line loss, the phase angle difference at both ends of the device, the change in the power flow through the device, and the change in the voltage amplitude at both ends of the device as the input of the neural network, the error parameters of the device are accurately identified. The steps are as follows: S1: Modify different parameters of the device through multiple experiments to generate errors, and collect the line loss, phase angle difference, power flow, and voltage amplitude changes caused by different errors. S2: Normalize the change amounts obtained through step S1. S3: Extract training samples and test samples from the characteristic quantities obtained through step S2. S4: Construct a BP neural network. S5: Train the BP neural network until a satisfactory accuracy is achieved. S6: Test the BP neural network.

3. The method for locating multi - point error parameters of a power flow calculation model based on stochastic hill - climbing according to claim 2, characterized in that: In step S3, the training sample data are respectively the normalized data of the change amounts of line loss, phase angle difference, power flow, and voltage amplitude.

4. The method for locating multi - point error parameters of a power flow calculation model based on stochastic hill - climbing according to claim 3, characterized in that: The method for data normalization is as follows: Let the power reference value in the power grid be SB and the voltage reference value be UB.

5. The method for locating multi - point error parameters of a power flow calculation model based on stochastic hill - climbing according to claim 3, characterized in that: The correspondence table between the training samples and the error parameter types is as follows: If the code is 1000, the error parameter type is resistance error; if the code is 0100, the error parameter type is inductance error; if the code is 0010, the error parameter type is susceptance error; if the code is 0001, the error parameter type is transformation ratio error. If there are multi-parameter errors, in the code, the corresponding positions are 1 and the rest are 0.

6. The method for locating multi - point error parameters of a power flow calculation model based on stochastic hill - climbing according to claim 4, characterized in that:The BP neural network described above is a feedforward BP neural network with one hidden layer. Among them, the input layer has 5 neurons, which respectively correspond to the change amount data in the training samples, and the output layer has 4 neurons, which respectively correspond to the error parameter types. The number of neurons in the hidden layer is set to 8.

7. The method for locating multi-point error parameters of the power flow calculation model based on random ramp climbing according to claim 5, wherein: Fuzzify the neural network output data, and set the data greater than 0.5 to 1 and the rest to 0.

8. The method for locating multi-point error parameters of the power flow calculation model based on random ramp climbing according to claim 6, wherein: Training the neural network according to the sample data is carried out using the trainlam function, the number of training times is 1000 times, and the learning rate is set to 0.

005.

9. The method for locating multi-point error parameters of the power flow calculation model based on random ramp climbing according to claim 1, wherein: In step four described above: Randomly locate several power devices. The method is as follows: Number the nodes within all error regions G from 1 to n. According to the different numbers of nodes in each error region, divide the region Gi into mi small regions on average. Randomly select one node in each small region, then nodes are randomly selected from the entire error region G. Let the numbers of these nodes be: a = {a1, a2,... aM}; Let the number of nodes in the region Gi be si, and mi = [0.1si], rounded down; Select the nodes numbered with the elements in the array a and start the positioning search. The method for starting the positioning search from the nodes is as follows: First, find all the lines connected to the node, including transformers, compare the power errors of these lines, select the line with the largest error as the positioning device, and conduct the next search for this positioning device. First, find the connected devices of the positioning device and conduct error comparison. If there are connected devices with an error larger than the positioning device, select the connected device with the largest error as the next positioning device. Finally, all the positioning searches will converge to some fixed positioning devices, and these devices are confirmed as error devices.

Images