Grid-connected frequency modulation control strategy based on photovoltaic power prediction

Abstract

The invention provides a grid-connected frequency modulation control strategy based on photovoltaic power prediction so as to solve the problems of stability and safety of a power system under high-proportion photovoltaic grid connection. According to the strategy, new parameters representing characteristics of a traditional synchronous generator set and a photovoltaic system are introduced, and a control strategy of a power grid side inverter is improved based on actual grid-connected working conditions. Meanwhile, a frequency dynamic response model considering the photovoltaic penetration influence is established, and the inertia response of the synchronous generator set is accurately simulated by introducing the dynamic characteristics of a power grid side direct current capacitor. And by adjusting the virtual inertia and the adjustment factor, the anti-disturbance capability of the photovoltaic system is enhanced. In combination with historical photovoltaic power generation data, the photovoltaic power is predicted by adopting data normalization, feature selection, SOM clustering analysis, singular spectrum analysis and a full-connection network model, and the prediction accuracy is improved. And the control strategy is combined with the power prediction result to realize efficient and stable frequency modulation control of the grid-connected photovoltaic system.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation and grid-connected frequency regulation control, specifically a grid-connected frequency regulation control strategy based on photovoltaic power prediction. Background Technology

[0002] With the rapid development of renewable energy, the penetration rate of photovoltaic (PV) power generation systems in power systems is gradually increasing. However, traditional power systems mainly rely on synchronous generators to provide inertial response and frequency stability, while PV systems, lacking rotational inertia, cannot provide natural frequency support like traditional generators. This characteristic leads to a significant reduction in the equivalent inertia of the power grid under high PV penetration, increasing the system frequency change rate and steady-state deviation, and may even cause frequency instability, seriously threatening the safe operation of the power system. Therefore, considering the grid-connected characteristics of PV systems, it is particularly important to study how to enhance the PV system's ability to support the grid frequency through virtual inertial control and frequency response optimization strategies. Simultaneously, PV power generation exhibits significant randomness and intermittency, and its output power is greatly affected by environmental factors, bringing greater uncertainty to the dynamic response of the grid frequency. Therefore, utilizing historical PV power generation data and improving the accuracy of PV power prediction through data analysis to provide basic data for formulating efficient grid-connected frequency regulation control strategies has become an important direction for solving current technical bottlenecks. To address the above problems, this invention proposes a grid-connected frequency regulation control strategy based on PV power prediction to improve the impact of PV systems on the dynamic characteristics of the grid frequency after grid connection and enhance the stability and security of the grid. Summary of the Invention

[0003] This invention proposes a grid-connected frequency regulation control strategy based on photovoltaic power prediction, which aims to solve the problem of deteriorating grid frequency stability under high photovoltaic penetration, so as to improve the impact of photovoltaic system on grid frequency dynamic characteristics after grid connection and improve grid stability and security.

[0004] To address the above problems, the present invention provides the following technical solution:

[0005] This paper analyzes the impact of grid-connected photovoltaic (PV) systems on the dynamic characteristics of grid frequency and proposes an improved method that introduces parameters such as proportional coefficients into the traditional grid frequency model to adapt to the increase in PV penetration, thereby more accurately simulating and predicting the frequency response of the grid under different PV penetration rates.

[0006] Under grid-connected conditions of photovoltaic systems, the frequency response characteristics of synchronous generator sets are simulated by introducing virtual inertia and adjustment factors, thereby improving the system's ability to resist frequency fluctuations caused by power imbalance.

[0007] The inertial response of a synchronous generator is simulated using grid-side DC capacitor dynamics to achieve virtual inertial control during grid-connected photovoltaic arrays. The virtual inertia provided by the grid-side DC capacitor mitigates the impact of photovoltaic grid connection on grid frequency stability.

[0008] By using historical photovoltaic power generation data, and through data normalization, feature selection, SOM cluster analysis, singular spectrum analysis, and fully connected network models, the accuracy of photovoltaic power prediction can be improved.

[0009] Furthermore, the impact of photovoltaic system grid connection on the dynamic characteristics of grid frequency is analyzed, and an improved method is proposed to introduce parameters such as proportional coefficient into the traditional grid frequency model to adapt to the increase in photovoltaic penetration rate, so as to more accurately simulate and predict the frequency response of the grid under different photovoltaic penetration rates.

[0010] When a photovoltaic system is connected to the grid, the traditional grid frequency model needs to introduce a parameter to describe the share of the traditional synchronous generator set or the photovoltaic system, which can be expressed as:

[0011]

[0012] In the formula, r represents the proportional coefficient of conventional synchronous generator sets, and the photovoltaic system penetration rate is 1-r, where 0≤r<1.

[0013] After introducing the proportional coefficient r into a conventional power grid closed-loop system, the transfer function from load disturbance to frequency disturbance is:

[0014]

[0015] In the formula, R is the governor adjustment factor, T is the equivalent inertia time constant of the prime mover, ξ is the damping ratio, and ω n Let H be the undamped oscillation frequency, H be the inertial time constant of the synchronous generator set, and F be the frequency of the undamped oscillation. HP G0 is the proportion of work done by the high-pressure cylinder of the prime mover, s is the Laplace variable, and z0 is the reciprocal of the prime mover reheat time constant.

[0016] Furthermore, under grid-connected conditions of photovoltaic systems, virtual inertia and adjustment factors are introduced to simulate the frequency response characteristics of synchronous generator sets, thereby improving the system's ability to resist frequency fluctuations caused by power imbalance.

[0017] According to the photovoltaic characteristic curve of the photovoltaic array, when the output voltage is less than the maximum power point, and there is an imbalance between the power output of the synchronous generator set and the load power, the system frequency will fluctuate significantly.

[0018] The inertial response of the synchronous generator set is linearized by dynamically simulating the DC capacitor on the grid side.

[0019]

[0020] In the formula, H dc U is the virtual inertia time constant, f is the nominal system frequency of 50Hz, f0 is the rated frequency of the power grid, and U dc0 The initial voltage of the DC-side capacitor is Δf; the change in grid frequency is ΔU. dc denoted as , where is the change in DC capacitor voltage, and k is the adjustment factor.

[0021] Adjustment factor k and virtual inertia time constant H dc These are important parameters affecting the control strategy, namely the adjustment factor k and the virtual inertia time constant H. dc Incremental DC voltage ΔU on the grid-connected side of the inverter dc Proportional. When the adjustment factor k and the virtual inertia time constant H dc When it increases, ΔU dc The more power a photovoltaic power station can provide to the system, the greater the supporting power it can provide, which can reduce fluctuations in the system's frequency dynamics.

[0022] Adding the transfer function link G(s) to the traditional power grid frequency response model, the transfer function G1(s) from load disturbance to frequency disturbance is as follows:

[0023]

[0024] Compared to traditional power grid frequency response models, the essence of this frequency regulation control strategy is to change the damping factor D of the synchronous generator set. The new damping factor becomes D+ak. f k.

[0025] Furthermore, the inertial response of a synchronous generator is simulated using the dynamics of a grid-side DC capacitor to achieve virtual inertial control when the photovoltaic array is connected to the grid. The virtual inertia provided by the grid-side DC capacitor is utilized to mitigate the impact of photovoltaic grid connection on grid frequency stability.

[0026] When the output voltage is greater than the maximum power point, the photovoltaic array operates to the right of the maximum power point. By superimposing the capacitor response power and the virtual inertial power, the total power change is formed to achieve nonlinear frequency stability control.

[0027] The inertial response of a synchronous generator set is simulated using the dynamics of a grid-side DC capacitor, i.e.:

[0028] ΔP cap =C dc2 U dc0 sΔU dc =C dc2 U dc0 skΔf

[0029] In the formula, ΔPcap This represents the capacitor's response power.

[0030] The DC voltage on the photovoltaic array side remains constant, so the output power of the photovoltaic array remains at a steady-state value. This process can be viewed as a result of virtual inertial control, therefore:

[0031]

[0032] In the formula, ΔP vir For virtual inertia control power, H PV This represents the virtual inertia of photovoltaics.

[0033] The change in total system power is:

[0034] ΔP total =ΔP cap +ΔP vir

[0035] The added virtual inertia corresponds to the superposition of the two processes mentioned above, thus adding a transfer function link to the traditional power grid frequency response model. The transfer function of this power superposition link is:

[0036]

[0037] The closed-loop transfer function of the system containing virtual control is updated as follows:

[0038]

[0039] The essence of this control strategy is to change the inertial time constant H of the synchronous generator set. The new virtual inertial time constant becomes:

[0040] H' dc =2H+kC dc2 U dc0 +2H PV

[0041] In the formula, H' dc This is the new virtual inertia time constant.

[0042] Furthermore, by using historical photovoltaic power generation data, and through data normalization, feature selection, SOM cluster analysis, singular spectrum analysis, and fully connected network models, the accuracy of photovoltaic power prediction can be improved.

[0043] Using historical photovoltaic power generation data, the input data x represents the i-th variable at time t. To avoid the adverse effects of different dimensions, each variable is normalized to [0,1], i.e.:

[0044]

[0045] In the formula, x′ is the normalized value, x is the original data, and x' is the normalized value. max x represents the maximum value of the data. min This represents the minimum value of the data.

[0046] Feature selection can effectively select input variables that are closely related to photovoltaic power, avoid interference from irrelevant variables on the model's power prediction, and reduce computational complexity.

[0047] The SOM clustering algorithm is a single-layer neural network based on competitive learning. Using the SOM self-organizing clustering method, photovoltaic data can be classified into three types: sunny days, cloudy / rainy days, and sudden change days.

[0048] By enhancing the network's influence on photovoltaic power fluctuation characteristics and performing SOM network calculations, the winning node can be identified.

[0049] Calculate the input vector X and the neuron node weights W ij Euclidean distance (t):

[0050]

[0051] In the formula, d j Let X be the Euclidean distance between the input vector and the activated neuron j. i W is the i-th component of the vector. ij (t) represents the weight between node i and node j.

[0052] Determine the winning node W c :

[0053] ||XW c ||=min{d j}

[0054] Update the weight vector:

[0055] W ij (t+1)=W ij (t)+η(t)h c,j (t)(X i -W ij (t))

[0056] In the formula, η(t) is the gain function, and h c,j (t) is a domain function.

[0057] Singular spectral analysis was used to decompose each cluster of data to obtain the main signal and residual signal.

[0058] The fully connected network uses the main signal and residual signal obtained from singular spectral decomposition as input. The fully connected network consists of multiple layers of neurons, and the calculation formula for each layer is as follows:

[0059] y = σ(WX + b)

[0060] In the formula, W is the weight matrix, b is the bias term, and σ is the activation function.

[0061] The predicted output power of the last layer of a fully connected network:

[0062] P pred =W out h f +b out

[0063] In the formula, P pred W is the predicted value. out h is the weight matrix of the output layer. f For the final output of the hidden layer, b out This is the bias term for the output layer.

[0064] Use mean squared error as the loss function to optimize network parameters:

[0065]

[0066] In the formula, J is the loss function value, and P i pred Let P be the predicted value for the i-th sample. i true Let be the true value of the i-th sample.

[0067] The fully connected network learns the nonlinear relationship between the main signal and the residual signal and generates a photovoltaic power prediction value for network output.

[0068] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0069] This invention improves the traditional frequency response model by quantifying the impact of photovoltaic penetration rate on grid inertia, significantly enhancing the prediction accuracy of system frequency stability under high-proportion photovoltaic grid integration and effectively suppressing frequency fluctuations and steady-state deviations. A virtual inertia dual-mode frequency regulation mechanism based on DC capacitor characteristics is proposed: enhancing the system's equivalent damping in the linear operating region and superimposing virtual inertia control to increase equivalent inertia in the nonlinear operating region, overcoming the limitations of existing technologies that rely on synchronous generators, significantly reducing the risk of frequency drops and shortening dynamic process recovery time. Combining weather clustering analysis and signal decomposition techniques, an intelligent photovoltaic power prediction model is constructed to provide high-precision forward-looking commands for virtual inertia control, solving the response lag problem of traditional methods. Stability theory proves that the control strategy can ensure the power angle converges within the safe operating range, significantly expanding the system stability margin, supporting a higher proportion of photovoltaic grid connection, while reducing frequency regulation reserve requirements and improving the economic efficiency and safety of grid operation. This strategy comprehensively optimizes frequency dynamic response characteristics and steady-state recovery capabilities, providing effective stable operation guarantees for high-proportion renewable energy grids. Attached Figure Description

[0070] Figure 1 The diagram shows a flowchart of a grid-connected frequency regulation control strategy based on photovoltaic power prediction.

[0071] Figure 2 The diagram shown illustrates the photovoltaic power prediction process. Detailed Implementation

[0072] S1 analyzes the impact of photovoltaic system grid connection on the dynamic characteristics of grid frequency, and proposes an improved method to introduce parameters such as proportional coefficient into the traditional grid frequency model to adapt to the increase of photovoltaic penetration rate, so as to more accurately simulate and predict the frequency response of the grid under different photovoltaic penetration rates.

[0073] S11, When a photovoltaic system is connected to the grid, the traditional grid frequency model needs to introduce a parameter to describe the share of the traditional synchronous generator set or the photovoltaic system, which can be expressed as:

[0074]

[0075] In the formula, r represents the proportional coefficient of a conventional synchronous generator set, and the photovoltaic system penetration rate is 1-r, where 0≤r<1. If the system load power is used as the reference value, the rotor time constant of the synchronous device and the output power of the turbine should both be multiplied by the coefficient r.

[0076] S12, After introducing the proportional coefficient r into the conventional power grid closed-loop system, the transfer function from load disturbance to frequency disturbance is:

[0077]

[0078] In the formula, R is the governor adjustment factor, T is the equivalent inertia time constant of the prime mover, ξ is the damping ratio, and ω n Let H be the undamped oscillation frequency, H be the inertial time constant of the synchronous generator set, and F be the frequency of the undamped oscillation. HP G0 is the proportion of work done by the high-pressure cylinder of the prime mover, s is the Laplace variable, and z0 is the reciprocal of the prime mover reheat time constant.

[0079] S13, When a step disturbance occurs in the load of the system, the grid frequency in the frequency domain is:

[0080]

[0081] In the formula, f r_pu (s) is the power grid frequency reference, f ref_pu (s) is ω represents the frequency interference value. d The inherent damping frequency.

[0082] The grid frequency in the time domain can be obtained through the inverse Laplace transform:

[0083]

[0084] in:

[0085]

[0086] z1 = 1 / T

[0087]

[0088] In the formula, ω d Z is the damped oscillation frequency, z1 is the disturbance intensity, A is the oscillation amplitude, and β is the phase offset.

[0089] As the penetration rate of photovoltaic systems continues to increase, the static photovoltaic modules lead to a decrease in the equivalent inertia of the system, which will gradually deteriorate the frequency characteristics provided to the power grid, increase the frequency change rate of the grid, increase the frequency steady-state deviation, and reduce the frequency minimum point, seriously endangering the safe operation of the power system.

[0090] S2, under the condition of grid connection of photovoltaic system, simulates the frequency response characteristics of synchronous generator set by introducing virtual inertia and adjustment factor, so as to improve the system’s anti-interference ability against frequency fluctuations caused by power imbalance.

[0091] S21. According to the photovoltaic characteristic curve of the photovoltaic array, when the output voltage is less than the maximum power point, and there is an imbalance between the power output of the synchronous generator set and the load power, the system frequency will fluctuate significantly.

[0092] Under the influence of unbalanced power, the equation of motion for the rotor in a synchronous generator set is:

[0093]

[0094] In the formula, δ is the rotor operating angle, ω0 is the rated rotor angular velocity, ω is the rotor electrical angular velocity, and H... SG P is the rotor inertia time constant. m P is the mechanical input power of the generator. e This refers to the electrical output power of the generator.

[0095] S22, the dynamic equation of the DC capacitor on the grid-connected side of the photovoltaic inverter is:

[0096]

[0097] In the formula, P dc For DC-side input power, P g C represents the grid-side output power. dc2 U is the capacitance of a DC capacitor. dc DC side voltage This represents the rate of change of the DC-side voltage.

[0098] S23, By dynamically simulating the DC capacitor on the grid side, the inertial response of the synchronous generator set can be obtained:

[0099]

[0100] In the formula, H dc U is the virtual inertia time constant, f is the nominal system frequency of 50Hz, f0 is the rated frequency of the power grid, and U dc0 This is the initial voltage of the DC-side capacitor.

[0101] Linearization yields:

[0102]

[0103] In the formula, Δf is the change in the power grid frequency, and ΔU dc denoted as , where is the change in DC capacitor voltage, and k is the adjustment factor.

[0104] Adjustment factor k and virtual inertia time constant H dc These are important parameters affecting the control strategy, namely the adjustment factor k and the virtual inertia time constant H. dc Incremental DC voltage ΔU on the grid-connected side of the inverter dc Proportional. When the adjustment factor k and the virtual inertia time constant H dc When it increases, ΔU dc The more power a photovoltaic power station can provide to the system, the greater the supporting power it can provide, which can reduce fluctuations in the system's frequency dynamics.

[0105] S24, The output power is obtained using a linear fitting method:

[0106] P = aV PV +b

[0107] In the formula, V PV Let be the output voltage of the photovoltaic system, and a and b be the fitting coefficients.

[0108] Linearization yields:

[0109] ΔP=aΔV PV =a·k f ΔU dc =a·k f kΔf

[0110]

[0111] In the formula, ΔP is the power change of the photovoltaic system, and k f It is a regulating factor.

[0112] Adding the transfer function link G(s) to the traditional power grid frequency response model, the transfer function G1(s) from load disturbance to frequency disturbance is as follows:

[0113]

[0114] Therefore, compared with the traditional power grid frequency response model, the essence of this frequency regulation control strategy is to change the damping factor D of the synchronous generator set. The new damping factor becomes D+ak. f k.

[0115] S3 uses grid-side DC capacitor dynamics to simulate the inertial response of a synchronous generator, achieving virtual inertial control when the photovoltaic array is connected to the grid. The grid-side DC capacitor provides virtual inertia, mitigating the impact of photovoltaic grid connection on grid frequency stability.

[0116] S31. According to the photovoltaic characteristic curve of the photovoltaic array, when the output voltage is greater than the maximum power point, the photovoltaic array is running on the right side of the maximum power point. It is impossible to linearly fit the right half. Therefore, by superimposing the capacitor response power and the virtual inertial power, the total power change is formed to achieve nonlinear frequency stability control.

[0117] The inertial response of a synchronous generator set is simulated using the dynamics of a grid-side DC capacitor, i.e.:

[0118] ΔP cap =C dc2 U dc0 sΔU dc =C dc2 U dc0 skΔf

[0119] In the formula, ΔP cap This represents the capacitor's response power.

[0120] The DC voltage on the photovoltaic array side remains constant, so the output power of the photovoltaic array remains at a steady-state value. This process can be viewed as a result of virtual inertial control, therefore:

[0121]

[0122] In the formula, ΔP vir For virtual inertia control power, H PV This represents the virtual inertia of photovoltaics.

[0123] The change in total system power is:

[0124] ΔP total =ΔP cap +ΔP vir

[0125] S32, the added virtual inertia corresponds to the superposition of the two processes mentioned above, thus adding a transfer function link to the traditional power grid frequency response model. The transfer function of this power superposition link is:

[0126]

[0127] The closed-loop transfer function of the system containing virtual control is updated as follows:

[0128]

[0129] The essence of this control strategy is to change the inertial time constant H of the synchronous generator set. The new virtual inertial time constant becomes:

[0130] H' dc =2H+kC dc2 U dc0 +2H PV

[0131] In the formula, H' dc This is the new virtual inertia time constant.

[0132] The resulting power system model is as follows:

[0133]

[0134] In the formula, δ g Let Δδ be the angle of effort. g This represents the change in the generator's power angle. U is the rate of change of the work angle. g U is the generator terminal voltage. t Let ΔP be the grid terminal voltage. mThis represents the change in mechanical power.

[0135] S33, the virtual inertial control of photovoltaics will affect the system's equivalent inertial time constant H and damping factor D. The Lyapunov energy function can be constructed as follows:

[0136]

[0137] In the formula, For work angular velocity, E k E is the kinetic energy of the system. p U is the system's potential energy, E0 is the system's initial potential energy, and U is the system's potential energy. g U is the generator terminal voltage. t X is the voltage of the power grid bus. g This refers to the reactor between the generator and the power grid.

[0138] The derivation of the Lyapunov energy function yields:

[0139]

[0140] in:

[0141]

[0142] In the formula, Let ω be the angular acceleration, and γ be the dynamic damping compensation term.

[0143] The rate of change of the Lyapunov function is:

[0144]

[0145] The attractive domain of a system can be estimated by using the energy at the unstable equilibrium point as a boundary value:

[0146]

[0147] In the formula, Ω L For the stable operation safety domain of the power system in the power angle-frequency phase plane, L th Let Lyapunov be the energy value of the system at the unstable equilibrium point.

[0148] S4 uses historical photovoltaic power generation data and improves the accuracy of photovoltaic power prediction through data normalization, feature selection, SOM cluster analysis, singular spectrum analysis, and fully connected network models.

[0149] S41 uses historical photovoltaic power generation data. Input data x represents the i-th variable at time t. To avoid adverse effects from different dimensions, each variable is normalized to [0,1], i.e.:

[0150]

[0151] In the formula, x′ is the normalized value, x is the original data, and x' is the normalized value. max x represents the maximum value of the data. min This represents the minimum value of the data.

[0152] S42, Feature selection can effectively select input variables that are closely related to photovoltaic power, avoid interference from irrelevant variables on the model's power prediction, and reduce computational complexity.

[0153] The Pearson correlation coefficient is used to measure the correlation between each variable and photovoltaic power. The formula for the Pearson correlation coefficient is:

[0154]

[0155] In the formula, x i y i For sample data, For the sample mean, in actual calculations, ρ1 represents the Pearson correlation coefficient between solar irradiance and photovoltaic module temperature, and ρ2 represents the Pearson correlation coefficient between solar irradiance and photovoltaic power.

[0156] S43, the SOM clustering algorithm, is a single-layer neural network based on competitive learning. Using the SOM self-organizing clustering method, it can classify photovoltaic data into three types: sunny days, cloudy / rainy days, and sudden change days.

[0157] By enhancing the network's influence on photovoltaic power fluctuation characteristics and performing SOM network calculations, the winning node can be identified.

[0158] Calculate the input vector X and the neuron node weights W ij Euclidean distance (t):

[0159]

[0160] In the formula, d j Let X be the Euclidean distance between the input vector and the activated neuron j. i W is the i-th component of the vector. ij (t) represents the weight between node i and node j.

[0161] Determine the winning node W c :

[0162] ||XW c ||=min{d j}

[0163] Update the weight vector:

[0164] W ij(t+1)=W ij (t)+η(t)h c,j (t)(X i -W ij (t))

[0165] In the formula, η(t) is the gain function, and h c,j (t) is a domain function, that is:

[0166]

[0167] In the formula, r(t) is a time-varying parameter representing the neighborhood radius.

[0168] S44, Apply singular spectral analysis to preprocess the time series data, transforming the one-dimensional time series x = {x} i |i=1,2,...,N} is embedded into the trajectory matrix M, that is:

[0169]

[0170] In the formula, N is the sequence length, L is the window length, K is the computation length, and K = N - L + 1.

[0171] Singular spectral analysis was used to decompose the data of each cluster to obtain the main signal and residual signal. The main signal represents the main trend of photovoltaic power, while the residual signal represents the high-frequency fluctuations and random perturbations of photovoltaic power.

[0172] S45, the fully connected network uses the main signal and residual signal obtained from singular spectral decomposition as inputs to the fully connected network. The fully connected network consists of multiple layers of neurons, and the calculation formula for each layer is:

[0173] y = σ(WX + b)

[0174] In the formula, W is the weight matrix, b is the bias term, and σ is the activation function.

[0175] The predicted output power of the last layer of a fully connected network:

[0176] P pred =W out h f +b out

[0177] In the formula, P pred W is the predicted value. out h is the weight matrix of the output layer. f For the final output of the hidden layer, b out This is the bias term for the output layer.

[0178] Use mean squared error as the loss function to optimize network parameters:

[0179]

[0180] In the formula, J is the loss function value, and P i pred Let P be the predicted value for the i-th sample. i true Let be the true value of the i-th sample.

[0181] The fully connected network learns the nonlinear relationship between the main signal and the residual signal and generates a photovoltaic power prediction value for network output.

Claims

1. A grid-connected frequency regulation control strategy based on photovoltaic power prediction, characterized in that, New parameters describing traditional synchronous generator sets and photovoltaic systems were applied. Based on actual grid connection conditions, the control strategy of the grid-side inverter was improved, and the mechanism of the frequency dynamic response model under photovoltaic penetration was analyzed. The model, by considering the dynamic characteristics of the grid-side DC capacitor, more accurately simulates the inertial response of the synchronous generator set. By adjusting the virtual inertia and regulation factor, the disturbance rejection capability of the photovoltaic system is enhanced. The control strategy, combined with historical photovoltaic power generation data, uses data normalization, feature selection, SOM cluster analysis, singular spectrum analysis, and a fully connected network model to predict photovoltaic power, thereby improving the accuracy of prediction. The control strategy method includes: S1. The impact of grid connection of photovoltaic system on the dynamic characteristics of grid frequency is analyzed, and an improved method is proposed to introduce parameters such as proportional coefficient into the traditional grid frequency model to adapt to the increase of photovoltaic penetration rate, so as to more accurately simulate and predict the frequency response of grid under different photovoltaic penetration rates. S2, under the condition of grid connection of photovoltaic system, virtual inertia and adjustment factor are introduced to simulate the frequency response characteristics of synchronous generator set, so as to improve the system’s anti-interference ability against frequency fluctuations caused by power imbalance. S3 uses grid-side DC capacitor dynamics to simulate the inertial response of a synchronous generator, realizing virtual inertial control when the photovoltaic array is connected to the grid; the grid-side DC capacitor provides virtual inertia, improving the impact of photovoltaic grid connection on grid frequency stability; S4 uses historical photovoltaic power generation data and improves the accuracy of photovoltaic power prediction through data normalization, feature selection, SOM cluster analysis, singular spectrum analysis, and fully connected network models.

2. The grid-connected frequency regulation control strategy based on photovoltaic power prediction according to claim 1, characterized in that, When a photovoltaic system is connected to the grid, the traditional grid frequency model needs to introduce a parameter to describe the share of the traditional synchronous generator set or the photovoltaic system, which can be expressed as: In the formula, r represents the proportional coefficient of conventional synchronous generator sets, and the photovoltaic system penetration rate is 1-r, where 0≤r<1.

3. The grid-connected frequency regulation control strategy based on photovoltaic power prediction according to claim 2, characterized in that, After introducing the proportional coefficient r into a conventional power grid closed-loop system, the transfer function from load disturbance to frequency disturbance is: In the formula, R is the governor adjustment factor, T is the equivalent inertia time constant of the prime mover, ξ is the damping ratio, and ω n Let H be the undamped oscillation frequency, H be the inertial time constant of the synchronous generator set, and F be the frequency of the undamped oscillation. HP G0 is the proportion of work done by the high-pressure cylinder of the prime mover, s is the Laplace variable, and z0 is the reciprocal of the prime mover reheat time constant.

4. The grid-connected frequency regulation control strategy based on photovoltaic power prediction according to claim 3, characterized in that, Under grid-connected conditions of photovoltaic systems, the frequency response characteristics of synchronous generator sets are simulated by introducing virtual inertia and adjustment factors, thereby improving the system's ability to resist frequency fluctuations caused by power imbalance. According to the photovoltaic characteristic curve of the photovoltaic array, when the output voltage is less than the maximum power point, and there is an imbalance between the power output of the synchronous generator set and the load power, the system frequency will fluctuate significantly. The inertial response of the synchronous generator set is linearized by dynamically simulating the DC capacitor on the grid side. In the formula, H dc U is the virtual inertia time constant, f is the nominal system frequency of 50Hz, f0 is the rated frequency of the power grid, and U dc0 The initial voltage of the DC-side capacitor is Δf; the change in grid frequency is ΔU. dc denoted as the change in DC capacitor voltage, and k is the adjustment factor. Adjustment factor k and virtual inertia time constant H dc These are important parameters affecting the control strategy, namely the adjustment factor k and the virtual inertia time constant H. dc Incremental DC voltage ΔU on the grid-connected side of the inverter dc Proportional; when the adjustment factor k and the virtual inertia time constant H dc When it increases, ΔU dc The more the photovoltaic power station increases, the greater the supporting power it can provide to the system, which can reduce fluctuations in the system's frequency dynamics. Adding the transfer function link G(s) to the traditional power grid frequency response model, the transfer function G1(s) from load disturbance to frequency disturbance is as follows: Compared to traditional power grid frequency response models, the essence of this frequency regulation control strategy is to change the damping factor D of the synchronous generator set. The new damping factor becomes D+ak. f k.

5. The grid-connected frequency regulation control strategy based on photovoltaic power prediction according to claim 4, characterized in that, The inertial response of a synchronous generator is simulated using grid-side DC capacitor dynamics to achieve virtual inertial control when the photovoltaic array is connected to the grid; the virtual inertia provided by the grid-side DC capacitor is used to improve the impact of photovoltaic grid connection on grid frequency stability. When the output voltage is greater than the maximum power point, the photovoltaic array operates to the right of the maximum power point. By superimposing the capacitor response power and the virtual inertial power, the total power change is formed to achieve frequency stability control in the nonlinear region. The inertial response of a synchronous generator set is simulated using the dynamics of a grid-side DC capacitor, i.e.: ΔP cap =C dc2 U dc0 sΔU dc =C dc2 U dc0 skΔf In the formula, ΔP cap This refers to the capacitor's response power. The DC voltage on the photovoltaic array side remains constant, so the output power of the photovoltaic array remains at a steady-state value. This process can be regarded as the result of virtual inertial control, therefore: In the formula, ΔP vir For virtual inertia control power, H PV For photovoltaic virtual inertia; The change in total system power is: ΔP total =ΔP cap +ΔP vir The added virtual inertia corresponds to the superposition of the two processes mentioned above, thus adding a transfer function link to the traditional power grid frequency response model. The transfer function of this power superposition link is: The closed-loop transfer function of the system containing virtual control is updated as follows: The essence of this control strategy is to change the inertial time constant H of the synchronous generator set. The new virtual inertial time constant becomes: H' dc =2H+kC dc2 U dc0 +2H PV In the formula, H' dc This is the new virtual inertia time constant.

6. The grid-connected frequency regulation control strategy based on photovoltaic power prediction according to claim 5, characterized in that, By using historical photovoltaic power generation data, we can improve the accuracy of photovoltaic power prediction through data normalization, feature selection, SOM cluster analysis, singular spectrum analysis, and fully connected network models. Using historical photovoltaic power generation data, the input data x represents the i-th variable of the time t data. To avoid the adverse effects of different dimensions, each variable is normalized to [0,1], that is: In the formula, x′ is the normalized value, x is the original data, and x' is the normalized value. max x represents the maximum value of the data. min The minimum value of the data; Feature selection can effectively select input variables that are closely related to photovoltaic power, avoid interference from irrelevant variables on the model's power prediction, and reduce computational complexity.

7. The grid-connected frequency regulation control strategy based on photovoltaic power prediction according to claim 6, characterized in that, The SOM clustering algorithm is a single-layer neural network based on competitive learning. Using the SOM self-organizing clustering method, photovoltaic data can be classified into three types: sunny days, cloudy / rainy days, and sudden change days. By enhancing the network's influence on photovoltaic power fluctuation characteristics and performing SOM network calculations, the winning node can be identified. Calculate the input vector X and the neuron node weights W ij Euclidean distance (t): In the formula, d j Let X be the Euclidean distance between the input vector and the activated neuron j. i W is the i-th component of the vector. ij (t) represents the weight between node i and node j. Determine the winning node W c : ||X-W c ||=min{d j } Update the weight vector: W ij (t+1)=W ij (t)+η(t)h c,j (t)(X i -W ij (t)) In the formula, η(t) is the gain function, and h c,j (t) is a domain function; Singular spectral analysis was used to decompose each cluster of data to obtain the main signal and residual signal.

8. The grid-connected frequency regulation control strategy based on photovoltaic power prediction according to claim 7, characterized in that, The fully connected network uses the main signal and residual signal obtained from singular spectral decomposition as inputs; the fully connected network consists of multiple layers of neurons, and the calculation formula for each layer is as follows: y = σ(WX + b) In the formula, W is the weight matrix, b is the bias term, and σ is the activation function; The predicted output power of the last layer of a fully connected network: P pred =W out h f +b out In the formula, P pred W is the predicted value. out h is the weight matrix of the output layer. f For the final output of the hidden layer, b out For the bias term of the output layer; Use mean squared error as the loss function to optimize network parameters: In the formula, J is the loss function value, and P i pred Let P be the predicted value for the i-th sample. i true This represents the true value of the i-th sample. The fully connected network learns the nonlinear relationship between the main signal and the residual signal and generates a photovoltaic power prediction value for network output.

Images