A method for calculating nonlinear time-varying transient reactance parameters of a large-capacity synchronous condenser based on BP neural network
By constructing a dataset on the relationship between current signals and reactance parameters and training a neural network model based on a BP neural network, the accuracy problem of reactance parameter calculation during the dynamic operation of large-capacity synchronous condensers was solved, and accurate description of reactance parameters and dynamic reactive power compensation were achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HARBIN UNIV OF SCI & TECH
- Filing Date
- 2023-02-28
- Publication Date
- 2026-05-26
AI Technical Summary
Existing methods for calculating transient reactance parameters cannot accurately describe the nonlinear magnetic saturation changes of large-capacity synchronous condensers during dynamic operation, resulting in inaccurate dynamic reactive power compensation characteristics.
A BP neural network-based method is adopted to construct a dataset of the relationship between current signals and reactance parameters by collecting stator and rotor currents and core permeability of synchronous condensers. The current variables are then trained using a BP neural network model as a reference standard for magnetic saturation, thereby achieving accurate calculation of time-varying transient reactance parameters.
It improves the calculation accuracy of reactance parameters during transient operation of motors, accurately describes the reactance changes of motors under different magnetic saturation levels, and supports dynamic reactive power compensation for large-capacity synchronous condensers.
Smart Images
Figure CN116861131B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of synchronous condenser technology, and in particular relates to a method for calculating the nonlinear time-varying transient reactance parameters of a large-capacity synchronous condenser based on a BP neural network. Background Technology
[0002] Large-capacity synchronous condensers, with their advantages of large reactive power and strong transient response, are used in my country's ultra-high-voltage direct current (UHVDC) transmission grid, characterized by "strong DC and weak AC," to address grid voltage stability issues. Compared to traditional synchronous condensers, large-capacity synchronous condensers, in addition to providing steady-state reactive power compensation under normal operating conditions, play a more crucial role in providing dynamic reactive power compensation, primarily under transient conditions, when voltage instability faults occur in the grid system. During dynamic reactive power compensation, the drastic changes in inrush current cause a high degree of magnetic saturation in the electromagnetic structure of the large-capacity synchronous condenser, resulting in nonlinear reactance parameters that change with the operation. Transient reactance parameters are crucial for describing the motor's operating characteristics, and changes in these parameters affect the dynamic reactive power compensation characteristics of the large-capacity synchronous condenser. Therefore, accurate calculation of transient reactance parameters is a prerequisite for accurate dynamic operation analysis of large-capacity synchronous condensers.
[0003] Commonly used methods for calculating transient reactance parameters include frequency domain identification and time domain identification. However, these methods are applicable to conventional synchronous motors operating in steady state and cannot meet the needs of dynamic operation analysis for large-capacity synchronous condensers. Therefore, in order to accurately describe the dynamic operating characteristics of large-capacity synchronous condensers, it is urgent to accurately calculate the nonlinear magnetic saturation changes of transient reactance parameters during the dynamic operation of large-capacity synchronous condensers. Summary of the Invention
[0004] This invention addresses the problem that existing conventional methods for calculating transient reactance parameters of synchronous motors can only characterize the nonlinear magnetic saturation degree during steady-state operation, but cannot consider the nonlinearity of transient reactance parameters changing with magnetic saturation degree during dynamic operation of large-capacity synchronous condensers. A method for calculating nonlinear time-varying transient reactance parameters of large-capacity synchronous condensers based on BP neural networks is provided.
[0005] A method for calculating the nonlinear time-varying transient reactance parameters of a large-capacity synchronous condenser based on a BP neural network includes the following steps:
[0006] Step 1: Calculate the transient field model of the synchronous condenser under no-load three-phase sudden short-circuit fault conditions. When the axis of the stator A-phase winding coincides with the rotor direct axis and the rotor quadrature axis respectively, collect and save the stator current, rotor current, and the corresponding relative permeability of the stator and rotor cores, including the stator A-phase current i. a Stator B-phase current i b Stator C-phase current ic Rotor current i f Relative permeability μ of the iron core r ;
[0007] Step 2: Assign the relative permeabilities of the stator and rotor cores saved in Step 1 to the constant field model and time-harmonic field model of the synchronous condenser, respectively, and calculate different reactance parameters:
[0008] First, calculate the synchronous reactance parameters. Assign the relative permeabilities of the stator and rotor cores saved in step one to the stator and rotor core positions in the constant field model of the synchronous condenser, respectively. Set the excitation winding to open circuit. The formula for calculating the synchronous reactance is:
[0009]
[0010] Among them, X d and X q These are the direct-axis synchronous reactance and quadrature-axis synchronous reactance of the synchronous condenser, respectively, ψ d and ψ q These are the direct-axis flux linkage and the quadrature-axis flux linkage, respectively, where ω is the angular frequency, and X... σ For stator end leakage reactance;
[0011] Next, the transient reactance parameters are calculated. The relative permeabilities of the stator and rotor cores saved in step one are assigned to the stator and rotor core positions in the eddy current field model of the synchronous condenser, respectively. The excitation winding is set to short-circuit, while the damping winding, rotor slot wedge, and rotor core are set to open circuit and eddy currents are ignored. The formula for calculating the transient reactance is as follows:
[0012]
[0013] Where, X′ d and X′ q These are the direct-axis transient reactance and quadrature-axis transient reactance of the synchronous modulator, respectively, ψ′ d and ψ′ q These are the instantaneous values of the direct-axis flux linkage and the quadrature-axis flux linkage, respectively.
[0014] Finally, the ultra-transient reactance parameters are calculated. The relative permeabilities of the stator and rotor cores saved in step one are assigned to the stator and rotor core positions in the eddy current field model of the synchronous condenser, respectively. The excitation winding is set to short-circuit, and the damping winding, rotor slot wedge, and rotor core are set to short-circuit. Taking eddy currents into account, the calculation formula for the ultra-transient reactance is:
[0015]
[0016] Where, X″ d and X″ q These are the direct-axis and quadrature-axis ultratransient reactances of the synchronous condenser, respectively, ψ″ d and ψ″ qThese are the instantaneous values of the direct-axis flux linkage and the quadrature-axis flux linkage, respectively.
[0017] Step 3: Standardize the stator and rotor signals saved in Step 1, and convert the stator current signal i... a i b and i c After transformation to the dq0 coordinate system, it is converted into the stator direct-axis current signal i. d and stator quadrature axis current signal i q ;
[0018] Step 4: Based on the reactance parameters and current signals obtained in Step 2 and Step 3, construct a dataset showing the correspondence between the stator and rotor current signals of the synchronous condenser and the saturation characteristics of different time-varying transient reactance parameters;
[0019] Step 5: Using the current variable as a reference standard for magnetic saturation, establish a neural network model with different time-varying transient reactance parameters. Input the dataset of the relationship between the current signal and the saturation characteristics of different time-varying transient reactance parameters constructed in Step 4 into the BP neural network model for training. Specifically:
[0020] The neural network model for different time-varying transient reactance parameters consists of an input layer, a hidden layer, and an output layer. The input layer has 3 neurons, the hidden layer has 19 neurons, and the output layer has 1 neuron. The input signal of the input layer is i. d i q and i f The output layer signal is a time-varying transient reactance parameter, the activation function of the hidden layer neurons is the Tansig function, and the activation function of the output layer neurons is the Purelin function.
[0021] Step Six: Determine whether the calculation error between the output reactance parameter of the neural network model and the corresponding actual reactance parameter is less than the expected error. If yes, output the time-varying transient reactance parameter; otherwise, proceed to Step Seven. The expected error is 0.001, and the calculation error is:
[0022]
[0023] Where e is the calculation error, and X is the output time-varying transient reactance parameter, X s These are the actual transient reactance parameters;
[0024] Step 7: Based on the calculation error in Step 6, adjust the weights and bias coefficients of the neural network models with different reactance parameters in reverse order to obtain the trained time-varying transient reactance parameter neural network model, specifically as follows:
[0025] The computational error is adjusted sequentially along the negative gradient direction, affecting the weights and biases of the output, hidden, and input layers. The adjustment formula is as follows:
[0026]
[0027] Where η is the learning rate (0.01), λ is the threshold of the output layer, and v k β is the weight of the k-th connection between the output layer and the hidden layer neurons. k w is the threshold of the k-th neuron in the hidden layer. jk The connection weights are the j-th neuron in the input layer and the k-th neuron in the hidden layer.
[0028] When the calculation error is less than the expected error, the adjustment of weights and bias coefficients ends, and a well-trained neural network model of time-varying transient reactance parameters is obtained.
[0029] The beneficial effects of this invention are:
[0030] This invention uses current variables as a reference standard for magnetic saturation and combines a BP neural network to characterize the nonlinear functional relationship between time-varying transient reactance parameters and multidimensional current variables. This realizes the change of reactance parameters with magnetic saturation during transient operation of the motor, thus improving the accuracy of parameter calculation. Attached Figure Description
[0031] Figure 1 This is a flowchart of a method for calculating nonlinear time-varying transient reactance parameters of a large-capacity synchronous condenser based on a BP neural network, as described in this invention.
[0032] Figure 2 This is a schematic diagram of the transient field simulation model of the synchronous condenser under no-load three-phase sudden short-circuit fault condition in step one of the specific implementation methods.
[0033] Figure 3 This describes the calculation method for transient reactance parameters at different times in step two of the specific implementation method;
[0034] Figure 4 This is a neural network structure diagram of the time-varying transient reactance parameters in step five of the specific implementation method. Detailed Implementation
[0035] Reference Figures 1 to 4 This embodiment describes a method for calculating the nonlinear time-varying transient reactance parameters of a large-capacity synchronous condenser based on a BP neural network, comprising the following steps:
[0036] Step 1: Establish a transient field simulation model of the synchronous condenser under no-load three-phase sudden short-circuit fault condition, and collect and save the stator current, rotor current, and the corresponding relative permeability of the stator and rotor cores, including the stator A-phase current i. a Stator B-phase current i b Stator C-phase current i c Rotor current if Relative permeability μ of the iron core r .
[0037] This embodiment collects and saves the stator current, rotor current signals, and corresponding relative permeability of the stator and rotor cores within 0-3 seconds after a sudden three-phase short-circuit fault occurs in the synchronous condenser. The sampling time is the moment when the axis of the stator A-phase winding coincides with the direct axis and quadrature axis of the rotor, respectively. Considering that the current change after a sudden three-phase short-circuit fault in the synchronous condenser can reach more than 10 times the rated current, and then gradually decreases to a steady state during dynamic operation, the magnetic saturation of the corresponding core gradually decreases to an unsaturated state, taking 0-3 seconds. Therefore, this embodiment collects the current signal and relative permeability of the core within 3 seconds after a sudden three-phase short-circuit fault occurs in the synchronous condenser.
[0038] Step Two: First, calculate the synchronous reactance parameters. Assign the relative permeabilities of the stator and rotor cores saved in Step One to the stator and rotor core positions in the constant field model of the synchronous condenser, respectively. Set the excitation winding to open circuit. The formula for calculating the synchronous reactance is:
[0039]
[0040] Among them, X d and X q These are the direct-axis synchronous reactance and quadrature-axis synchronous reactance of the synchronous condenser, respectively, ψ d and ψ q These are the direct-axis flux linkage and the quadrature-axis flux linkage, respectively, where ω is the angular frequency, and X... σ For stator end leakage reactance;
[0041] Next, the transient reactance parameters are calculated. The relative permeabilities of the stator and rotor cores saved in step one are assigned to the stator and rotor core positions in the eddy current field model of the synchronous condenser, respectively. The excitation winding is set to short-circuit, while the damping winding, rotor slot wedge, and rotor core are set to open circuit and eddy currents are ignored. The formula for calculating the transient reactance is as follows:
[0042]
[0043] Where, X′ d and X′ q These are the direct-axis transient reactance and quadrature-axis transient reactance of the synchronous modulator, respectively, ψ′ d and ψ′ q These are the instantaneous values of the direct-axis flux linkage and the quadrature-axis flux linkage, respectively.
[0044] Finally, the ultra-transient reactance parameters are calculated. The relative permeabilities of the stator and rotor cores saved in step one are assigned to the stator and rotor core positions in the eddy current field model of the synchronous condenser, respectively. The excitation winding is set to short-circuit, and the damping winding, rotor slot wedge, and rotor core are set to short-circuit. Taking eddy currents into account, the calculation formula for the ultra-transient reactance is:
[0045]
[0046] Where, X″ d and X″ q These are the direct-axis and quadrature-axis ultratransient reactances of the synchronous condenser, respectively, ψ″ d and ψ″ q These are the instantaneous values of the direct-axis flux linkage and the quadrature-axis flux linkage, respectively.
[0047] The transient field model of the synchronous condenser, the stator current signal, the rotor current signal, the relative permeability of the stator core, the relative permeability of the rotor core, and the constant field model and eddy current field model of the synchronous condenser described in step one above, as well as the constant field model and eddy current field model of the synchronous condenser described in step two above, are all obtained through finite element simulation software, such as Flux simulation software.
[0048] Step 3: Standardize the stator and rotor signals saved in Step 1, and convert the stator current signal i... A i B and i C After transformation to the dq0 coordinate system, it is converted into the stator direct-axis current signal i. d and stator quadrature axis current signal i q .
[0049] Step 4: Based on the reactance parameters and current signals obtained in Step 2 and Step 3, construct a dataset of the correspondence between the stator and rotor current signals of the synchronous condenser and the saturation characteristics of different reactance parameters.
[0050] Step 5: Using the current variable as a reference standard for magnetic saturation, establish a neural network model with different time-varying transient reactance parameters. Input the dataset of the relationship between the current signal and the saturation characteristics of different time-varying transient reactance parameters constructed in Step 4 into the BP neural network model for training. Specifically:
[0051] The neural network model for different time-varying transient reactance parameters consists of an input layer, a hidden layer, and an output layer. The input layer has 3 neurons, the hidden layer has 19 neurons, and the output layer has 1 neuron. The input signal of the input layer is i. d i q and i f The output layer signal is a time-varying transient reactance parameter, the activation function of the hidden layer neurons is the Tansig function, and the activation function of the output layer neurons is the Purelin function. The neural network model structure is as follows: Figure 3As shown, where ∑ k ∑ represents the sum of the products of the connection weights of the k-th hidden layer neuron and the input signals of the input layer neuron, and ∑ represents the sum of the products of the output signals of the hidden layer neuron and the connection weights of the output layer.
[0052] Step Six: Determine whether the calculation error between the output reactance parameter of the neural network model and the corresponding actual reactance parameter is less than the expected error. If yes, output the time-varying transient reactance parameter; otherwise, proceed to Step Seven. The expected error is 0.001, and the calculation error is:
[0053]
[0054] Where e is the calculation error, and X is the output time-varying transient reactance parameter, X s These are the actual transient reactance parameters.
[0055] Step 7: Based on the calculation error in Step 6, adjust the weights and bias coefficients of the neural network models with different reactance parameters in reverse order to obtain the trained time-varying transient reactance parameter neural network model, specifically as follows:
[0056] The computational error is adjusted sequentially along the negative gradient direction, affecting the weights and biases of the output, hidden, and input layers. The adjustment formula is as follows:
[0057]
[0058] Where η is the learning rate (0.01), λ is the threshold of the output layer, and v k β is the weight of the k-th connection between the output layer and the hidden layer neurons. k w is the threshold of the k-th neuron in the hidden layer. jk The weights are the connection weights between the j-th neuron in the input layer and the k-th neuron in the hidden layer. When the computational error is less than the expected error, the adjustment of the weights and bias coefficients ends, resulting in a well-trained neural network model with time-varying transient reactance parameters.
[0059] This implementation method, based on the calculation results of a sudden three-phase short-circuit fault under no-load conditions in a large-capacity synchronous condenser, collects current signals and the relative permeability of the iron core to calculate transient reactance parameters at different times, i.e., different degrees of saturation. It constructs a dataset showing the correspondence between the stator and rotor current signals of the synchronous condenser and the saturation characteristics of different transient reactance parameters. Based on this dataset, a neural network model with different time-varying transient reactance parameters is trained. The obtained neural network model, using the current variable as the reference standard for the degree of saturation, can accurately describe the nonlinear magnetic saturation changes of transient reactance parameters during the dynamic operation of the synchronous condenser.
Claims
1. A method for calculating time-varying transient reactance parameters of a large-capacity synchronous condenser based on a BP neural network, characterized in that, Includes the following steps: Step 1: Establish a transient field model of the synchronous condenser to calculate the no-load three-phase sudden short-circuit fault condition. Collect and save the stator current, rotor current, and corresponding relative permeability of the stator and rotor cores at different times, including the stator A-phase current i. a Stator B-phase current i b Stator C-phase current i c Rotor current i f Relative permeability μ of the iron core r ; Step 2: Assign the relative permeability of the stator and rotor cores saved in Step 1 to the constant field model and time-harmonic field model of the synchronous condenser, respectively, and calculate different reactance parameters; Step 3: Standardize the stator and rotor signals saved in Step 1, and convert the stator current signal i... a i b and i c After transformation to the dq0 coordinate system, it is converted into the stator direct-axis current signal i. d and stator quadrature axis current signal i q ; Step 4: Based on the reactance parameters and current signals obtained in Step 2 and Step 3, construct a dataset showing the correspondence between the stator and rotor current signals of the synchronous condenser and the saturation characteristics of different reactance parameters; Step 5: Using the current variable as a reference standard for the degree of magnetic saturation, establish a neural network model with different time-varying transient reactance parameters. Input the dataset of the relationship between the current signal and the saturation characteristics of different reactance parameters constructed in Step 4 into the BP neural network model for training. Step Six: Determine whether the calculation error between the output reactance parameter of the neural network model and the corresponding actual reactance parameter is less than the expected error. If yes, output the time-varying transient reactance parameter; otherwise, proceed to Step Seven. Step 7: Based on the calculation error in Step 6, adjust the weights and bias coefficients of the neural network models with different reactance parameters in reverse order to obtain the trained time-varying transient reactance parameter neural network model.
2. The method for calculating time-varying transient reactance parameters of a large-capacity synchronous condenser based on a BP neural network according to claim 1, characterized in that, In step one, the stator current signal, rotor current signal, and relative permeability of the stator and rotor cores are collected within 0 to 3 seconds after the synchronous condenser suddenly short-circuits under no-load three-phase conditions. The collection time is the moment when the axis of the stator A-phase winding coincides with the rotor direct axis and the rotor quadrature axis, respectively.
3. The method for calculating time-varying transient reactance parameters of a large-capacity synchronous condenser based on a BP neural network according to claim 1, characterized in that, Step two is as follows: First, assign the relative permeabilities of the stator and rotor cores saved in step one to the stator and rotor core positions in the constant field model of the synchronous condenser, respectively. Set the excitation winding to open circuit, and calculate the synchronous reactance using the following formula: Among them, X d and X q These are the direct-axis synchronous reactance and quadrature-axis synchronous reactance of the synchronous condenser, respectively, ψ d and ψ q These are the direct-axis flux linkage and the quadrature-axis flux linkage, respectively, where ω is the angular frequency, and X... σ For stator end leakage reactance; Secondly, the relative permeabilities of the stator and rotor cores saved in step one are assigned to the stator and rotor core positions in the eddy current field model of the synchronous condenser, respectively. The excitation winding is set to short-circuit, and the damping winding, rotor slot wedge, and rotor core are set to open circuit and eddy currents are ignored. The formula for calculating the transient reactance is: Where, X′ d and X′ q These are the direct-axis transient reactance and quadrature-axis transient reactance of the synchronous modulator, respectively, ψ′ d and ψ′ q These are the instantaneous values of the direct-axis flux linkage and the quadrature-axis flux linkage, respectively. Finally, the relative permeabilities of the stator and rotor cores saved in step one are assigned to the stator and rotor core positions in the eddy current field model of the synchronous condenser, respectively. The excitation winding is set to short-circuit, and the damping winding, rotor slot wedge, and rotor core are set to short-circuit. Taking eddy currents into account, the formula for calculating the super-transient reactance is: Where, X″ d and X″ q These are the direct-axis and quadrature-axis ultratransient reactances of the synchronous condenser, respectively, ψ″ d and ψ″ q These are the instantaneous values of the direct-axis flux linkage and the quadrature-axis flux linkage, respectively.
4. The method for calculating time-varying transient reactance parameters of a large-capacity synchronous condenser based on a BP neural network according to claim 1, characterized in that, The neural network model with different time-varying transient reactance parameters described in step five includes an input layer, a hidden layer, and an output layer. The input layer consists of 3 neurons, the hidden layer contains 19 neurons, and the output layer contains 1 neuron. The input signal of the input layer is i. d i q and i f The output layer signal is a time-varying transient reactance parameter.
5. The method for calculating time-varying transient reactance parameters of a large-capacity synchronous condenser based on a BP neural network according to claim 1, characterized in that, In the neural network models with different reactance parameters described in step five, the activation function of the hidden layer neurons is the Tansig function, and the activation function of the output layer neurons is the Purelin function.
6. The method for calculating time-varying transient reactance parameters of a large-capacity synchronous condenser based on a BP neural network according to claim 1, characterized in that, The expected error mentioned in step six is 0.001, and the calculated error is: Where e is the calculation error, and X is the output time-varying transient reactance parameter, X s These are the actual transient reactance parameters.
7. The method for calculating time-varying transient reactance parameters of a large-capacity synchronous condenser based on a BP neural network according to claim 1, characterized in that, Step seven specifically involves: calculating the error and adjusting the weights and biases of the output layer, hidden layer, and input layer sequentially in the direction of the negative gradient. The adjustment formula is as follows: Where η is the learning rate (0.01), λ is the threshold of the output layer, and v k β is the weight of the k-th connection between the output layer and the hidden layer neurons. k w is the threshold of the k-th neuron in the hidden layer. jk The connection weights are the j-th neuron in the input layer and the k-th neuron in the hidden layer. When the calculation error is less than the expected error, the adjustment of weights and bias coefficients ends, and a well-trained neural network model of time-varying transient reactance parameters is obtained.