A photovoltaic power station power generation amount prediction method and system based on multi-source data fusion

Abstract

The application relates to the technical field of smart grids, and discloses a photovoltaic power station power generation amount prediction method and system based on multi-source data fusion, which comprises the following steps: constructing a sensitivity matrix according to network frame admittance and section flow characteristics and generating a dispatch weight vector; parallelly checking a prediction expected sequence and a vector to establish a whole-network flow out-of-limit risk coefficient; calculating a discrete change rate of a power deviation amount according to grid-connected point flow power; adjusting a control period in combination with the risk coefficient and the change rate, determining a power compensation amount to adjust inverter cluster output, and eliminating control lag caused by independent prediction through time sequence remodeling, locking micro-disturbance in situ, effectively inhibiting inverter cluster control bounce and flow resonance, and improving flow safety boundary convergence precision.

Description

Technical Field

[0001] This invention belongs to the field of smart grid technology, and in particular relates to a method and system for predicting the power generation of photovoltaic power plants based on multi-source data fusion. Background Technology

[0002] Currently, when the power system performs active power dispatch, the forecast results of distributed generation are usually used as external input. A regression model is established based on historical output data and meteorological parameters to provide the dispatch master station with the expected output reference curve of active power. This is used to formulate the power balance plan of the whole network and maintain the stability of the power flow section of the grid. With the increase in the proportion of large-capacity distributed photovoltaic access, the meteorological conditions in different regions are becoming more unstable, and the active power at the photovoltaic grid connection point is more likely to fluctuate intermittently. Existing forecasting methods that rely on spatiotemporal statistical averages are insufficient to absorb short-term and rapid power flow impacts.

[0003] When meteorological conditions change abruptly, a time deviation will occur between the actual output and the predicted output of the photovoltaic substation grid connection point, injecting transient power disturbances into the grid. Current methods typically first predict power generation and then perform power flow section verification. The prediction and safety verification stages are cascaded in time, and the lack of dynamic coordination between the admittance parameters of each node and the local inverter control loop easily leads to communication delays between dispatch command issuance and inverter response, resulting in active power parameter divergence and power flow section oscillations. Operators often have to resort to conservative power limiting methods. To reduce the risk of power flow exceeding limits, conventional practices attempt to increase the data acquisition frequency or accelerate the prediction calculation frequency to compensate for the problem of untimely information updates. However, simply increasing the computing density can easily push industrial central processing units to near computing power saturation, while also increasing communication overhead and equipment costs. Besides the aforementioned hardware limitations such as processor computing power saturation and communication overhead, existing control methods... There are also shortcomings in dealing with the over-limit of the entire grid section. For example, Chinese invention patent application CN121484883A discloses a method, device, terminal equipment and storage medium for eliminating power flow over-limit based on section sensitivity and source load characteristics. Although it attempts to reduce the computational complexity by constructing a targeted sub-model through section pre-classification, this method implicitly relies on the underlying properties of the quasi-static distribution of source load characteristics. Faced with short-term and rapid transient power flow impacts under the high proportion of distributed photovoltaic grid-connected conditions, the scheme still cannot avoid control hysteresis due to the lack of time-series reshaping for prediction and safety verification, and the lack of adaptive feedback closed loop for the adjustment cycle of the inverter local control loop. It cannot effectively suppress the control bounce and power flow resonance of the inverter cluster. Since the system lacks an admittance feedback closed loop for grid impedance changes, the existing method still cannot suppress the frequent oscillation of the inverter cluster control loop, and it does not solve the problem of the disconnect between the predicted curve and the actual power flow constraints of the grid.

[0004] Therefore, the technical problem to be solved by this invention is how to integrate the influence of multi-source data changes and grid impedance on power flow distribution during the active power prediction process, reorganize the timing process of active power prediction and power flow section safety boundary verification, and achieve adaptive control and transient stability adjustment of active power output under different risk ranges. Summary of the Invention

[0005] This invention aims to solve the problem of reshaping power prediction and power flow section verification timing process to achieve adaptive cutoff control of active power output and transient voltage stability offset adjustment.

[0006] In this technical solution, a method for predicting the power generation of a photovoltaic power plant based on multi-source data fusion includes the following steps: Step S1: Obtain the topological admittance characteristic parameters of the network nodes, extract the active power flow direction characteristics of the tie line sections, construct the active power flow sensitivity matrix of the network nodes and generate the active power scheduling weight vector of the controlled nodes based on it. Step S2: The expected value sequence of active power prediction of photovoltaic power station and the active power scheduling weight vector of controlled node are introduced into the power flow safety verification logic in parallel, and the power flow section safety boundary is calculated simultaneously to establish the over-limit risk coefficient of the power flow section of the whole network. Step S3: Obtain the real-time power flow power of the grid-connected point during the current sampling period, calculate the active power deviation between it and the expected active power value sequence, and calculate the discrete first-order gradient change rate of the active power deviation within the sliding control window. Step S4: Determine the control interval and adjust the sampling control cycle value based on the absolute value of the over-limit risk coefficient of the entire network power flow section and the discrete first-order gradient change rate, calculate the active power compensation adjustment amount, and control the active power output state of the photovoltaic inverter cluster; wherein, when the absolute value of the change rate is less than the disturbance threshold for M consecutive cycles, the active power compensation adjustment amount is set to 0; when the over-limit risk coefficient of the entire network power flow section is within 0.75 to 0.90, the sampling control cycle value is maintained at the standard reference cycle value; when the over-limit risk coefficient of the entire network power flow section is greater than 0.90, the sampling control cycle value is shortened to 0.2 times it.

[0007] Preferably, step S4 includes the following sub-steps: Step S41, when the absolute value of the discrete first-order gradient change rate is less than the set micro-amplitude disturbance threshold in multiple consecutive sampling control cycle values, it is determined that the fluctuation of the real-time power flow at the grid connection point is in the low-frequency noise range of the active power flow; Step S42, the calibration value of the adaptive adjustment coefficient is set to 0, the active power compensation adjustment amount is kept at 0, and the photovoltaic inverter cluster is controlled to maintain the current output state.

[0008] Preferably, step S4 further includes the following sub-steps: step S43, when the risk coefficient of the power flow section exceeding the limit is greater than 0.90 and is within the second control range, shorten the sampling control cycle value; step S44, set the sampling control cycle value to 0.2 times the standard reference cycle value, and update the active power compensation adjustment amount according to the shortened sampling control cycle value.

[0009] Preferably, the method further includes the following steps: Step S5, obtain the real-time frequency value of the grid connection point, and when the real-time frequency value exceeds the set frequency limit threshold and the risk coefficient of the power flow section of the whole network exceeds the set safety threshold, stop adjusting the sampling control cycle value, and adjust the active power output state of the photovoltaic inverter cluster according to the set droop control characteristic curve.

[0010] Preferably, step S1 includes the following sub-steps: step S11, acquiring the interconnection impedance data of each node in the controlled grid and extracting the node admittance matrix feature values ​​of the tie line section under different power flow directions; step S12, calculating the power flow sensitivity coefficient of each controlled node to the power fluctuation of the tie line section based on the node admittance matrix feature values, and combining the power flow sensitivity coefficients to generate the grid node active power-section power flow sensitivity matrix.

[0011] Preferably, step S2 includes the following sub-steps: Step S21, calculate the singular value decomposition of the active power-section power flow sensitivity matrix of the network nodes to generate the active power scheduling weight vector of the controlled nodes; Step S22, introduce the expected value sequence of active power prediction and the active power scheduling weight vector of the controlled nodes in parallel into the power flow safety verification logic to calculate the power flow distribution state of the entire network nodes, and calculate the power flow section over-limit risk coefficient of the entire network based on the degree of deviation between the power flow distribution state and the network safety margin.

[0012] Preferably, step S3 includes the following sub-steps: step S31, setting a fixed number of sampling control cycle values ​​within the sliding control window; step S32, obtaining the active power deviation in the current time period, calculating the discrete difference of the active power deviation within multiple consecutive sampling control cycle values, so as to generate a discrete first-order rate of change.

[0013] Preferably, step S4 includes the following sub-steps: step S45, mapping the active power compensation adjustment amount to the power allocation coefficient of each photovoltaic inverter node, and sending it to the corresponding photovoltaic inverter node through the active power scheduling channel; step S46, adjusting the output power of each photovoltaic inverter node according to the power allocation coefficient to change the active power output state of the photovoltaic inverter cluster.

[0014] Preferably, before step S1, the following steps are also included: Step S0, acquiring historical meteorological observation data, historical power output data, and real-time operating data, normalizing and aligning the acquired data with time synchronization, so as to generate a sequence of expected active power values.

[0015] A photovoltaic power plant power generation prediction system based on multi-source data fusion, which is used to implement a photovoltaic power plant power generation prediction method based on multi-source data fusion, includes: The data feature purification module is used to acquire historical meteorological observation data, historical power output data, and real-time operating data, normalize and time-synchronize the acquired data to generate a sequence of expected active power prediction values. The network topology analysis module is used to obtain the topological admittance characteristic parameters of the network nodes, extract the active power flow direction characteristics of the tie line sections, construct the active power flow sensitivity matrix of the network nodes and generate the active power scheduling weight vector of the controlled nodes based on the active power flow sensitivity matrix of the network nodes. The power flow boundary verification module is used to introduce the expected value sequence of active power prediction and the active power scheduling weight vector of controlled nodes into the power flow safety verification logic in parallel, and simultaneously calculate the power flow section safety boundary to establish the risk coefficient of the power flow section exceeding the limit for the whole network. The transient deviation calculation module is used to obtain the real-time power flow power of the grid-connected point within the current sampling period, calculate the active power deviation between the real-time power flow power of the grid-connected point and the expected value sequence of active power prediction, and calculate the discrete first-order rate of change of the active power deviation within the sliding control window. The adaptive hedging scheduling control module is used to determine the control interval and adjust the sampling control cycle value based on the absolute value of the over-limit risk coefficient and the discrete first-order rate of change of the entire network power flow section, calculate the active power compensation adjustment amount, and control the active power output status of the photovoltaic inverter cluster.

[0016] Compared with existing technologies, the photovoltaic power generation prediction method and system based on multi-source data fusion of the present invention has the following advantages: 1. In the prediction of photovoltaic power generation by multi-source data fusion, by comparing the real-time power flow at the grid connection point with the expected value of active power prediction, and by identifying minor disturbances in combination with the changes in active power deviation, this invention can avoid frequent adjustments of photovoltaic inverter clusters due to minor fluctuations in weather conditions. When the deviation change is small, the active power compensation adjustment is kept at 0, which can reduce unnecessary control actions and keep the inverter cluster in a relatively stable output state, thereby reducing equipment losses and operational risks caused by repeated actions of the control loop.

[0017] 2. By introducing the active power-section power flow sensitivity matrix of the network nodes and the active power scheduling weight vector of the controlled nodes, this invention enables the expected value sequence of active power prediction to be simultaneously checked with the safety boundary of the power flow section while forming the scheduling basis. This can reduce the time lag caused by the traditional method of prediction first and then verification, enabling the scheduling side to identify the over-limit trend of the power flow section more timely, and improve the speed of judging the safety status of the entire network power flow and the timeliness of scheduling response.

[0018] 3. When the risk coefficient of the power flow section exceeding the limit increases, this invention shortens the sampling control cycle value, so that the active power compensation adjustment can be updated more quickly with the power deviation of the grid connection point. This method improves the active power response capability of the photovoltaic inverter cluster to heavy load or transient conditions, reduces the adjustment lag caused by long control cycle, helps to suppress power flow oscillation, and makes the convergence process of the power flow section safety boundary more stable. Attached Figure Description

[0019] Figure 1 This is a flowchart of the calculation process for network-wide over-limit risk and power deviation gradient in this invention; Figure 2 This is a diagram of the adaptive interlocking control structure for inverter cluster output in this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0021] A method for predicting the power generation of a photovoltaic power plant based on multi-source data fusion includes the following steps: Step S1: Obtain the topological admittance characteristic parameters of the network nodes, extract the active power flow direction characteristics of the tie line sections, construct the active power flow sensitivity matrix of the network nodes and generate the active power scheduling weight vector of the controlled nodes based on it. Step S2: The expected value sequence of active power prediction of photovoltaic power station and the active power scheduling weight vector of controlled node are introduced into the power flow safety verification logic in parallel, and the power flow section safety boundary is calculated simultaneously to establish the over-limit risk coefficient of the power flow section of the whole network. Step S3: Obtain the real-time power flow power of the grid-connected point during the current sampling period, calculate the active power deviation between it and the expected active power value sequence, and calculate the discrete first-order gradient change rate of the active power deviation within the sliding control window. Step S4: Determine the control interval and adjust the sampling control cycle value based on the absolute value of the over-limit risk coefficient of the entire network power flow section and the discrete first-order gradient change rate, calculate the active power compensation adjustment amount, and control the active power output state of the photovoltaic inverter cluster; wherein, when the absolute value of the change rate is less than the disturbance threshold for M consecutive cycles, the active power compensation adjustment amount is set to 0; when the over-limit risk coefficient of the entire network power flow section is within 0.75 to 0.90, the sampling control cycle value is maintained at the standard reference cycle value; when the over-limit risk coefficient of the entire network power flow section is greater than 0.90, the sampling control cycle value is shortened to 0.2 times it.

[0022] Preferably, step S4 includes the following sub-steps: Step S41, when the absolute value of the discrete first-order gradient change rate is less than the set micro-amplitude disturbance threshold in multiple consecutive sampling control cycle values, it is determined that the fluctuation of the real-time power flow at the grid connection point is in the low-frequency noise range of the active power flow; Step S42, the calibration value of the adaptive adjustment coefficient is set to 0, the active power compensation adjustment amount is kept at 0, and the photovoltaic inverter cluster is controlled to maintain the current output state.

[0023] Preferably, step S4 further includes the following sub-steps: step S43, when the risk coefficient of the power flow section exceeding the limit is greater than 0.90 and is within the second control range, shorten the sampling control cycle value; step S44, set the sampling control cycle value to 0.2 times the standard reference cycle value, and update the active power compensation adjustment amount according to the shortened sampling control cycle value.

[0024] Preferably, the method further includes the following steps: Step S5, obtain the real-time frequency value of the grid connection point, and when the real-time frequency value exceeds the set frequency limit threshold and the risk coefficient of the power flow section of the whole network exceeds the set safety threshold, stop adjusting the sampling control cycle value, and adjust the active power output state of the photovoltaic inverter cluster according to the set droop control characteristic curve.

[0025] Preferably, step S1 includes the following sub-steps: step S11, acquiring the interconnection impedance data of each node in the controlled grid and extracting the node admittance matrix feature values ​​of the tie line section under different power flow directions; step S12, calculating the power flow sensitivity coefficient of each controlled node to the power fluctuation of the tie line section based on the node admittance matrix feature values, and combining the power flow sensitivity coefficients to generate the grid node active power-section power flow sensitivity matrix.

[0026] Preferably, step S2 includes the following sub-steps: Step S21, calculate the singular value decomposition of the active power-section power flow sensitivity matrix of the network nodes to generate the active power scheduling weight vector of the controlled nodes; Step S22, introduce the expected value sequence of active power prediction and the active power scheduling weight vector of the controlled nodes in parallel into the power flow safety verification logic to calculate the power flow distribution state of the entire network nodes, and calculate the power flow section over-limit risk coefficient of the entire network based on the degree of deviation between the power flow distribution state and the network safety margin.

[0027] Preferably, step S3 includes the following sub-steps: step S31, setting a fixed number of sampling control cycle values ​​within the sliding control window; step S32, obtaining the active power deviation in the current time period, calculating the discrete difference of the active power deviation within multiple consecutive sampling control cycle values, so as to generate a discrete first-order rate of change.

[0028] Preferably, step S4 includes the following sub-steps: step S45, mapping the active power compensation adjustment amount to the power allocation coefficient of each photovoltaic inverter node, and sending it to the corresponding photovoltaic inverter node through the active power scheduling channel; step S46, adjusting the output power of each photovoltaic inverter node according to the power allocation coefficient to change the active power output state of the photovoltaic inverter cluster.

[0029] Preferably, before step S1, the following steps are also included: Step S0, acquiring historical meteorological observation data, historical power output data, and real-time operating data, normalizing and aligning the acquired data with time synchronization, so as to generate a sequence of expected active power values.

[0030] A photovoltaic power plant power generation prediction system based on multi-source data fusion, which is used to implement a photovoltaic power plant power generation prediction method based on multi-source data fusion, includes: The data feature purification module is used to acquire historical meteorological observation data, historical power output data, and real-time operating data, normalize and time-synchronize the acquired data to generate a sequence of expected active power prediction values. The network topology analysis module is used to obtain the topological admittance characteristic parameters of the network nodes, extract the active power flow direction characteristics of the tie line sections, construct the active power flow sensitivity matrix of the network nodes and generate the active power scheduling weight vector of the controlled nodes based on the active power flow sensitivity matrix of the network nodes. The power flow boundary verification module is used to introduce the expected value sequence of active power prediction and the active power scheduling weight vector of controlled nodes into the power flow safety verification logic in parallel, and simultaneously calculate the power flow section safety boundary to establish the risk coefficient of the power flow section exceeding the limit for the whole network. The transient deviation calculation module is used to obtain the real-time power flow power of the grid-connected point within the current sampling period, calculate the active power deviation between the real-time power flow power of the grid-connected point and the expected value sequence of active power prediction, and calculate the discrete first-order rate of change of the active power deviation within the sliding control window. The adaptive hedging scheduling control module is used to determine the control interval and adjust the sampling control cycle value based on the absolute value of the over-limit risk coefficient and the discrete first-order rate of change of the entire network power flow section, calculate the active power compensation adjustment amount, and control the active power output status of the photovoltaic inverter cluster.

[0031] Example 1: In this example, the photovoltaic power generation prediction method is used for the active power balance control of a smart grid with a high proportion of distributed photovoltaic access. When transient anomalies occur in the passing meteorological elements, the photovoltaic collection station grid connection point is prone to non-Gaussian distributed active power output residuals. The high-frequency fluctuations of these residuals will impact the steady-state power flow of the transmission section, causing active power imbalance at the grid nodes and transient deviation of the system frequency. If one-way open-loop data fitting is still used, the prediction response lag will cause the control rack to operate frequently, increasing the mechanical fatigue wear of the equipment.

[0032] To mitigate real-time power flow impacts at grid connection points, the grid status monitoring unit collects topological admittance characteristic parameters of the smart grid nodes under the current operating section and extracts the active power flow direction characteristics of the tie-line sections. Based on the topological admittance characteristic parameters and active power flow direction characteristics, the smart grid central dispatch processor constructs an active power flow sensitivity matrix between the grid nodes and the section, calculates the marginal impact of changes in active power output of each controlled node on the overall power flow distribution, and generates active power dispatch weight vectors for the controlled nodes. Simultaneously, the data allocation unit acquires meteorological sensor sequence parameters and historical output parameters of the inverter cluster, outputs a normalized prediction feature matrix after nonlinear alignment transformation, and forms an active power prediction expectation value sequence from this matrix.

[0033] After receiving the expected value sequence of active power prediction and the active power scheduling weight vector of controlled nodes, the power flow safety verification logic verifies the safety boundary of the power flow section in parallel to obtain the over-limit risk coefficient of the entire network power flow section. Specifically, the power flow safety verification logic is executed by the central scheduling processor of the intelligent network structure. It expands the expected value sequence of active power prediction according to the time section and performs a spatial matrix multiplication operation with the corresponding active power scheduling weight vector of controlled nodes to convert the active power prediction increment of each controlled node into the forward power flow active power estimate of the tie line section. Subsequently, the power flow active power estimate is compared with the static thermal stability power flow limit and transient stability limit power boundary of the tie line section stored in the register. The approximation degree of the power flow active power estimate to the power flow section safety boundary is calculated. That is, the power flow active power estimate is divided by the limit power boundary of the section to determine the over-limit risk coefficient of the entire network power flow section. The risk coefficient is a scalar between 0 and 1. The closer the value is to 1, the closer the current power flow state is to the over-limit boundary of the section.

[0034] The intelligent network structure central dispatch processor performs real-time maintenance, including a fixed number of [items / items]. A sliding control window with one sampling period is used to extract the real-time power flow of the grid-connected points within the current sampling window. With active power prediction expectation The real-time difference is used as the active power deviation, and the discrete first-order gradient rate of change of the active power deviation within the sliding control window is calculated. When the absolute value of the discrete first-order gradient rate of change is within the continuous... When the amplitude of the disturbance is less than the preset threshold within each sampling period, the intelligent grid central scheduling processor will adaptively adjust the coefficient. The calibration weights are cleared to zero, maintaining the active power compensation adjustment output at 0, thus controlling the photovoltaic inverter cluster to maintain its current active power output state. This represents the total number of sampling periods within the sliding control window. Real-time power flow at the grid connection point This represents the expected value of the predicted active power. The threshold for the continuous sampling period used to monitor the persistent state of the disturbance. This is the adaptive adjustment coefficient calibrated based on the active power-cross-section power flow sensitivity matrix of the grid nodes.

[0035] When the risk coefficient of the entire network power flow section exceeding the limit is within the first control range of 0.75 to 0.90, the sampling control cycle value is... Maintaining the standard baseline period value, the intelligent grid central dispatch processor corrects the expected active power value sequence based on the active power dispatch weight vector of the controlled nodes, and calculates the active power compensation adjustment amount. The specific calculation formula is as follows: ,in, This is the active power compensation adjustment amount. The adaptive adjustment coefficients are calibrated based on the active power-section power flow sensitivity matrix of the grid nodes. Real-time power flow at the grid connection point This represents the expected value of the predicted active power. The sampled control cycle value is used as a control step size adjustment factor in the discretization implementation of actual control engineering. It is used to dynamically scale the adjustment intensity within the current cycle and complete the dimensional balance through the control time constant reference. This ensures that the calculated active power compensation adjustment amount is directly mapped to the active power target deviation command that the photovoltaic inverter cluster should respond to within the current control cycle when it is sent to the inverter actuator.

[0036] When the risk coefficient of the entire network power flow section exceeding the limit is greater than 0.90 and enters the second control interval, the intelligent network central scheduling processor will sample the control cycle value. Switching to high-frequency reduced cycle, the value of which is set to 0.2 times the standard reference cycle, shortens the control response adjustment delay. In this state, if the real-time frequency value of the grid connection point exceeds the set frequency limit threshold, and the risk coefficient of the entire network power flow section exceeding the limit is greater than the set safety threshold, the intelligent grid central dispatch processor stops adjusting the sampling control cycle value and triggers the local frequency extreme value hardware action interlocking feedback gating. After the gating interlocks the digital dispatch command transmission loop, the photovoltaic inverter cluster adjusts the active power output state according to the set droop control characteristic curve. The droop control characteristic curve is used to characterize the correspondence between the deviation of the real-time frequency value of the grid connection point relative to the rated frequency and the active power output adjustment amount. The larger the frequency deviation, the larger the corresponding active power output adjustment amount.

[0037] To avoid control deadlock or signal blind spots during the switch from software digital computing chain to local hardware droop control, the grid-connected point local frequency extreme value hardware action interlocking feedback gating incorporates a high-speed logic linkage triggering mechanism. When the real-time frequency value of the grid-connected point is detected to exceed the set frequency limit threshold, the hardware gating executes a rigid physical interlocking action within 2 milliseconds, shutting down the digital bus channel. Simultaneously, the zero-order state holder configured on the end side freezes and latches the last frame of digital power dispatching instructions at the instant before interlocking, using this instruction as the underlying safety power reference. Subsequently, the hardware droop control loop, based on this reference, fine-tunes the active power output through a high-speed field-programmable gate array chip with a hardware response cycle of 15 milliseconds. When the active power flow of the grid-connected topology stabilizes and the real-time frequency value of the grid-connected point recovers to the preset safety range for 5 consecutive seconds, the hardware gating performs smooth unblocking and gradually restores the weight allocation of the digital dispatching instructions through a linear damping transition time window of 3 seconds, causing the control loop to close again.

[0038] Offline progressive active power compensation and adjustment commands generated by the central dispatch processor of the smart grid are transmitted to the inverter cluster control rack of the photovoltaic power station. The control rack adjusts the active power output of each inverter node according to these commands, smoothing out transient power surges caused by non-Gaussian distributed active power output residuals, and maintaining the frequency variation range of the smart grid aggregation station nodes within a preset ±0.2Hz adjustment range. This ensures a dynamic power distribution steady state between the active power prediction curve and the network topology active power flow constraints. During the specific mapping and allocation process, the photovoltaic power station inverter cluster control rack receives the total active power... After adjusting the power compensation amount, the current real-time remaining adjustable capacity of each photovoltaic inverter node in the cluster is obtained, which is the difference between the rated capacity of each inverter and the current actual output value. Then, a capacity ratio allocation strategy is adopted to multiply the total active power compensation adjustment amount by the proportion of the remaining adjustable capacity of each inverter node to the total remaining adjustable capacity of the cluster. The power allocation coefficient of each photovoltaic inverter node is calculated in this way. The sum of the power allocation coefficients of all nodes is 1. Finally, each inverter node adjusts its own active power output according to its obtained power allocation coefficient to achieve a coordinated response to the overall compensation adjustment command.

[0039] Example 2: In this example, the active power prediction and system scheduling control system's active power balance test is conducted on a digital simulation test bench that includes a high proportion of distributed photovoltaic access. The digital simulation test bench solves the power flow control equations of the power grid based on the physical admittance law of the grid topology impedance. It calls standard meteorological sensor sequence parameters from a publicly available meteorological dataset and historical output parameters of the inverter cluster as data input sources to reproduce the residual fluctuation of active power output at the substation and grid connection points during transient changes in passing meteorological elements. The measurement and transmission unit used for signal acquisition has a sampling rate of not less than 1000Hz, and the intelligent grid central scheduling processor has double-precision floating-point arithmetic capability. During operation, the digital simulation test bench actively superimposes Gaussian white noise with a signal-to-noise ratio of 20dB from the local signal source and introduces power frequency interference harmonics with a frequency of 50Hz to assess the dynamic power distribution physical balance between the predicted active power value and the active power flow constraints of the network topology.

[0040] Before the experiment started, the intelligent grid central scheduling processor determined the total number of sampling periods within the sliding control window through a data buffer reconstruction step within the discrete time window. With continuous sampling period threshold Total number of sampling periods This is used to balance the time-domain identification accuracy of transient signal waveform capture with the bus communication load of the central processing unit's scheduling and control flow; when the deviation signal fluctuation frequency between the expected value of the predicted active power at the grid connection point and the real-time power flow at the grid connection point increases and the spectral topology width widens, the total number of sampling periods... The value tends towards the upper limit of the range to reduce the impact of digital signal aliasing on the calculation of the discrete first-order gradient rate of change. Under typical high-frequency non-Gaussian output residual impact conditions, the total number of sampling periods is... Set to 10, continuous sampling period threshold Set to 3; Discrete-time gradient adaptive truncation control logic extracts the real-time power flow of the grid-connected points within the current sampling window. With active power prediction expectation The real-time difference is used as the active power deviation, and the discrete first-order gradient rate of change of the active power deviation within the sliding control window is calculated. When the absolute value of the discrete first-order gradient rate of change is within the continuous... When the amplitude of the disturbance is less than the preset threshold within each sampling period, the intelligent grid central scheduling processor will adaptively adjust the coefficient. The calibration weight is cleared to zero, so that the output of the active power compensation regulation is kept at 0, thereby shutting off the inverter disorder regulation feedback increment caused by random weather changes.

[0041] The parameters to consider include the risk coefficient of exceeding the limit at the entire network power flow section, the sampling control cycle value, and the adaptive adjustment coefficient. The frequency variation range of the aggregation station nodes was compared between the experimental group and the control group using a digital simulation test bench. Within the first control interval where the risk coefficient for exceeding the limit of the power flow section of the entire network was 0.82, the experimental group maintained a sampling control cycle value of 1.00s and an adaptive adjustment coefficient. The value is set to 1.45. The intelligent network central dispatch processor calls the network node active power-section power flow sensitivity matrix to solve the active power dispatch weight vector of the controlled node, and calculates the active power compensation adjustment amount accordingly. The frequency variation range of the collecting station node is 0.08Hz, and the steady-state equilibrium state is stable. Correspondingly, under the condition that the risk coefficient of the power flow section exceeding the limit of the entire network is 0.82 and the sampling control period value is 1.00s, the adaptive adjustment coefficient is set to 1.45. When set to 0.00, it is impossible to form an effective active power compensation adjustment based on the active power-cross-section power flow sensitivity matrix of the grid nodes. The frequency variation range of the collecting station nodes expands to 0.42Hz, and the steady-state balance state is out of limit.

[0042] When the risk coefficient of the entire network power flow section exceeding the limit rises to 0.94 and enters the second control interval, the experimental group shortens the sampling control period to 0.20s and adjusts the adaptive adjustment coefficient. Still taking 1.45; under the action of the shortened sampling control period, the active power compensation adjustment is updated in a timely manner according to the deviation between the real-time power flow at the grid connection point and the expected value of the active power prediction. The frequency variation range of the collection station node is 0.14Hz, and the steady-state balance remains stable. The control group 2 has the same risk coefficient of 0.94 and adaptive adjustment coefficient at the power flow section of the whole network. Under the same condition of 1.45, the sampling control period value of 1.00s is maintained. The discrete control flow generates a time delay, the frequency variation range of the aggregation station node increases to 0.38Hz, and the steady-state equilibrium state is out of limit.

[0043] When the risk coefficient deviates from the boundary of the first control interval, the data further reflects the impact of the matching relationship between the sampling control period value and the adaptive adjustment coefficient on the system steady state. The over-limit risk coefficient of the entire network power flow section in control group three is 0.72, which is lower than the lower limit of 0.75. The sampling control period value is 1.00s, and the adaptive adjustment coefficient... The value is 0.85, the frequency variation range of the aggregation station nodes is 0.28Hz, the steady-state equilibrium state shows fluctuation, the over-limit risk coefficient of the entire network power flow section of control group four is 0.96, which is higher than the upper limit of 0.90, the sampling control period value is 0.20s, and the adaptive adjustment coefficient is... The value is 2.65, the frequency variation range of the aggregation station node is 0.31Hz, and the steady-state equilibrium state is characterized by resonance.

[0044] Near the endpoint of the first control interval, the over-limit risk coefficient of the power flow section of the entire network at the lower boundary point of the test group is 0.75, the sampling control period is 1.00s, and the adaptive adjustment coefficient is... The value is 1.10, the frequency variation range of the aggregation station nodes is 0.11Hz, and the steady-state equilibrium state is stable; the risk coefficient of the entire network power flow section exceeding the limit at the upper boundary point of the test group is 0.90, the sampling control period is 1.00s, and the adaptive adjustment coefficient is... The value is 1.80, the frequency variation range of the aggregation station nodes is 0.15Hz, and the steady-state equilibrium state is also stable. After further increasing the risk level, the over-limit risk coefficient of the entire network power flow section at the extreme operating point of the test group is 0.98, the sampling control cycle value is switched to 0.20s, and the adaptive adjustment coefficient is... The value is 1.95, the frequency variation range of the aggregation station node is 0.19Hz, and the steady-state equilibrium remains stable.

[0045] In the aforementioned test data, when the over-limit risk coefficient of the entire network power flow section was within the range of 0.75 to 0.90, the test team calculated the active power compensation adjustment using a standard reference period of 1.00s and the active power scheduling weight vector of the controlled nodes, controlling the frequency variation range of the collecting station nodes within 0.08Hz to 0.15Hz. Under heavy load conditions where the over-limit risk coefficient of the entire network power flow section was greater than 0.90, the test team shortened the sampling control period to 0.20s, increased the active power response bandwidth of the converter control loop, and controlled the frequency variation range of the collecting station nodes within 0. From 0.14Hz to 0.19Hz, control group 1 exhibited a frequency variation range of 0.42Hz due to the lack of effective active power compensation regulation, while control group 2 exhibited a frequency variation range of 0.38Hz due to maintaining a sampling control period of 1.00s. Control group 3 showed continuous fluctuations at a state below the lower limit of 0.72, while control group 4 showed resonance at a state above the upper limit of 0.96. This indicates that the risk range of 0.75 to 0.90 and the period shortening rule when it is greater than 0.90 can jointly constrain the release rhythm of the active power compensation regulation.

[0046] In this embodiment, the offline progressive active power compensation and adjustment command generated by the central dispatch processor of the smart grid is transmitted to the photovoltaic power station inverter cluster control rack. The photovoltaic power station inverter cluster control rack adjusts the active power output of each inverter node according to the command, smooths the transient power impact caused by the non-Gaussian distribution of active power output residual, and maintains the frequency variation range of the smart grid collection station node within the preset ±0.2Hz adjustment range, so that the active power prediction curve and the network topology active power flow constraint maintain a dynamic power distribution steady state.

[0047] Example 3: When the active power jump occurs at the current power grid connection point and the communication and control link experiences a time delay, the active power output status of the photovoltaic inverter cluster is easily affected by the low-frequency drift interference measured in the collected voltage and power data. After this interference enters the data allocation unit, it will cause phase misalignment in the normalized prediction feature matrix and cause non-steady-state fluctuations in the over-limit risk coefficient of the whole network power flow section obtained by the power flow safety verification logic, thereby triggering frequent adjustment actions of the inverter converter control loop.

[0048] In this embodiment, the data allocation unit takes the collected high-frequency 1-second real-time operating data and low-frequency 15-minute meteorological radiance observation data as input, and uses a cubic spline nonlinear interpolation algorithm to smooth the temporal sequence of the meteorological data to align the meteorological data with the real-time operating data on the time axis. Subsequently, the maximum-minimum normalization method is used to map the aligned multi-source feature data to a dimensionless interval of 0 to 1, constructing a normalized prediction feature matrix with consistent time scale and standardized amplitude, and generating an active power prediction expectation value sequence based on this matrix for parallel calculation by the power flow safety verification logic.

[0049] To correct the impact of low-frequency drift interference on the risk coefficient of power flow sections exceeding limits across the entire network, the intelligent grid central dispatch processor reads the original power variation sequence collected by high-frequency power sensors, uses first-order differential filtering to remove the low-frequency slow drift component from the signal, and calculates the active power scheduling weight vector of controlled nodes based on the online-maintained active power-section power flow sensitivity matrix of grid nodes. This determines the marginal impact of active power output changes of each controlled node on the power flow distribution under the current operating section. The processed active power prediction expectation sequence and the active power scheduling weight vector of controlled nodes are then synchronously entered into the power flow safety verification logic, so that the risk coefficient of power flow sections exceeding limits across the entire network is updated according to the current operating status of the grid connection points.

[0050] When the risk coefficient of exceeding the limit at the power flow section of the entire network exceeds 0.75 and enters the first control interval, the central dispatch processor of the intelligent grid structure extracts the real-time power flow of the grid-connected points within the sliding control window online. With active power prediction expectation The real-time difference between the two values ​​is used as the active power deviation. The discrete first-order gradient rate of change of the active power deviation within the sliding control window is then calculated. When the absolute value of the discrete first-order gradient rate of change is within the continuous... When the amplitude of the disturbance is less than the preset threshold within each sampling period, the adaptive adjustment coefficient will be used. The calibration weights are reset to 0, maintaining the active power compensation adjustment. The output value is 0, which controls the photovoltaic inverter cluster to maintain the current active power output state, thereby limiting the inconsistency of data among multiple parties in the control bus communication network.

[0051] Sampling control cycle value Adjustments are made based on the over-limit risk coefficient of the entire power flow section, switching between the first and second control intervals. The intelligent grid central dispatch processor calculates and issues active power compensation adjustment amounts based on the adjusted sampled control cycle values. The photovoltaic inverter cluster adjusts the active power output of each inverter node to smooth out the transient power impact caused by the non-Gaussian distribution of active power output residuals, and controls the frequency variation range of the smart grid aggregation station nodes within the preset ±0.2Hz adjustment range, so that the active power prediction curve and the active power flow constraints of the network topology are kept in balance.

[0052] Example 4: This example combines Figures 1 to 2 This paper describes a method and system for predicting the power generation of photovoltaic power plants based on multi-source data fusion, such as... Figure 1As shown, step S1 obtains the topological admittance characteristic parameters of the grid nodes, extracts the active power flow direction characteristics of the tie-line sections, constructs the active power flow sensitivity matrix of the grid nodes and generates the active power scheduling weight vector of the controlled nodes based on it. Then, it proceeds to step S2, where the expected value sequence of the photovoltaic power plant's active power prediction and the active power scheduling weight vector of the controlled nodes are introduced into the power flow safety verification logic in parallel, and the safety boundary of the power flow section is calculated simultaneously to establish the risk coefficient of the power flow section exceeding the limit for the entire network. Then, it proceeds to step S3, obtains the real-time power flow of the grid-connected points in the current sampling period, calculates the active power deviation between it and the expected value sequence of the active power prediction, and calculates the active power deviation. The discrete first-order gradient change rate within the sliding control window is measured, and finally step S4 is executed. The control interval is determined based on the absolute value of the power flow section over-limit risk coefficient and the discrete first-order gradient change rate, and the sampling control cycle value is adjusted. The active power compensation adjustment amount is calculated to control the active power output state of the photovoltaic inverter cluster. When the absolute value of the change rate is less than the disturbance threshold for M consecutive cycles, the active power compensation adjustment amount is set to 0. When the power flow section over-limit risk coefficient is between 0.75 and 0.90, the sampling control cycle value is maintained at the standard reference cycle value. When the power flow section over-limit risk coefficient is greater than 0.90, the sampling control cycle value is shortened to 0.2 times it.

[0053] like Figure 2 As shown, the real-time power flow is collected and transferred to the synchronous calculation of the power flow section safety boundary through the input expected value and weight. When the power flow is continuously less than the micro-disturbance threshold, the power flow is transferred to maintain the current output state. In the processing of the synchronous calculation of the power flow section safety boundary, if it is in the first control interval, the power flow is transferred to maintain the standard reference period value. If it enters the second control interval, the power flow is transferred to shorten the sampling control period value. If the frequency limit threshold is exceeded after shortening the sampling control period value, the power flow is further transferred to execute the rigid physical blocking action. When the safety range is continuously restored, the power flow is transferred from executing the rigid physical blocking action to restoring the digital scheduling command weight. Finally, when the control loop is closed again, the power flow returns to the synchronous calculation of the power flow section safety boundary.

[0054] Example 5: When the distributed photovoltaic inverter cluster grid connection point is in the initial power supply and commissioning phase, the site first maintains an open-loop reactive power compensation state. The intelligent grid central dispatch processor continuously collects the real-time power flow data of the grid connection point for 72 hours through high-frequency power sensors. With active power prediction expectation The historical sequence is used to calculate the variance envelope of the steady-state deviation sequence of the two online, and the power background noise baseline value of the power grid is calculated accordingly. This baseline value is used to characterize the level of random power disturbance under the current grid topology environment. Its three times value is set as the small disturbance threshold in the discrete time gradient adaptive truncation control logic, which is used to determine whether meteorological random noise needs to trigger active power compensation adjustment.

[0055] During the same commissioning phase, the master station injects pulse disturbance power into the grid connection point and simultaneously records the change in the active power flow direction of the tie line section. The intelligent grid central dispatch processor uses the least squares fitting algorithm to calculate the initial marginal impact of the change in the admittance characteristic parameters of each node on the power flow distribution of the entire network, thereby completing the initial parameter filling of the active power flow sensitivity matrix of the grid node and establishing the control adaptive parameter benchmark for this grid topology environment.

[0056] After the aforementioned adaptive control parameter benchmark is imported, the active power balance control loop of the digital simulation test bench enters online operation. The data allocation unit acquires multi-source heterogeneous characteristic data of the output residuals accompanying the non-Gaussian distribution, and the power flow safety verification logic calculates the over-limit risk coefficient of the entire network power flow section in parallel. As the photovoltaic inverter cluster adjusts according to the active power compensation amount... The risk coefficient exhibits a monotonically decreasing trend as it is released in a stepped manner. When the simulated transient power flow at the transmission section crosses the critical point of the steady-state power flow safety boundary, the discrete-time gradient adaptive truncation control logic will sample the control period value. The response bandwidth of the active power control loop is increased by shortening it to 0.2 times the standard reference period.

[0057] After the sampling control period is shortened, the intelligent grid central dispatch processor adjusts the power flow based on the real-time power flow at the grid connection point. With active power prediction expectation The active power deviation between them, update the active power compensation adjustment. It drives the photovoltaic inverter cluster to adjust the active power output of each inverter node. As the active power response bandwidth of the converter control loop increases, the frequency transient deviation of the smart grid aggregation station node is smoothed out to a regulation fluctuation range of ±0.2Hz, so that the active power prediction curve within the smart grid topology safety boundary and the active power flow constraint of the network topology maintain a dynamic power distribution balance.

[0058] Example 6: During the long-term alternating operation of the power transmission network, the aging of components at the grid connection point of the current photovoltaic power plant will cause a physical shift in the topological admittance characteristic parameters. This shift changes the active power flow direction reference of the tie line section, causing the static initial value of the active power flow sensitivity matrix of the grid node to deviate from the actual operating state. The power flow safety verification logic will generate a systematic deviation in the risk coefficient of the entire network power flow section exceeding the limit, which will cause the active power compensation adjustment to lag or over-adjust.

[0059] To correct the topology admittance calculation bias caused by slow parameter drift, the intelligent grid central dispatch processor calls the cycle calibration program during the zero-load maintenance window. The grid status monitoring unit continuously collects topology admittance characteristic parameters and power flow sequence of transmission sections for 24 hours. The intelligent grid central dispatch processor recalculates the marginal impact of active power output changes of each controlled node on the power flow distribution of the entire network through its internal least squares parameter update loop, and writes the calibrated topology parameters into the register array of the active power-power flow sensitivity matrix of the grid nodes. At the same time, the data allocation unit extracts the historical output prediction bias variance over a preceding 30 days and adaptively corrects the normalization scaling factor of the normalized prediction feature matrix to prevent the input vector from exceeding the boundary limit of the normalized prediction feature matrix.

[0060] After the register array completes the parameter writing, the system resumes online scheduling and control of active power at the grid connection point. The corrected active power-section power flow sensitivity matrix of the grid node and the normalized prediction feature matrix operate in coordination at the substation measurement and control level to capture the non-Gaussian output residual caused by meteorological anomalies. The power flow safety verification logic parallel solves the output of the network power flow section over-limit risk coefficient, which maintains a monotonically convergent state with the actual power flow over-limit risk of the transmission section.

[0061] When the risk coefficient of the entire network power flow section exceeding the limit is greater than 0.90 and the system enters a heavy-load transient condition, the discrete-time gradient adaptive truncation control logic switches to a variable-period adjustment mode, and the sampled control period value is adjusted. The time is reduced to 0.2s. Based on this, the central dispatch processor of the smart grid updates the active power compensation adjustment amount and drives the photovoltaic inverter cluster control rack to adjust the active power output of each inverter node. This keeps the frequency transient deviation of the smart grid aggregation station node within the preset ±0.2Hz adjustment range, so that the active power prediction curve and the network topology active power flow constraint maintain a dynamic power distribution steady state.

[0062] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. A method for predicting the power generation of a photovoltaic power plant based on multi-source data fusion, characterized in that, Includes the following steps: Step S1: Obtain the topological admittance characteristic parameters of the network nodes, extract the active power flow direction characteristics of the tie line sections, construct the active power flow sensitivity matrix of the network nodes and generate the active power scheduling weight vector of the controlled nodes based on it. Step S2: The expected value sequence of active power prediction of photovoltaic power station and the active power scheduling weight vector of controlled node are introduced into the power flow safety verification logic in parallel, and the power flow section safety boundary is calculated simultaneously to establish the over-limit risk coefficient of the power flow section of the whole network. Step S3: Obtain the real-time power flow power of the grid-connected point during the current sampling period, calculate the active power deviation between it and the expected active power value sequence, and calculate the discrete first-order gradient change rate of the active power deviation within the sliding control window. Step S4: Determine the control interval and adjust the sampling control cycle value based on the absolute value of the over-limit risk coefficient of the entire network power flow section and the discrete first-order gradient change rate, calculate the active power compensation adjustment amount, and control the active power output state of the photovoltaic inverter cluster; wherein, when the absolute value of the change rate is less than the disturbance threshold for M consecutive cycles, the active power compensation adjustment amount is set to 0; when the over-limit risk coefficient of the entire network power flow section is within 0.75 to 0.90, the sampling control cycle value is maintained at the standard reference cycle value; when the over-limit risk coefficient of the entire network power flow section is greater than 0.90, the sampling control cycle value is shortened to 0.2 times it.

2. The photovoltaic power generation prediction method based on multi-source data fusion according to claim 1, characterized in that, Step S4 includes the following sub-steps: Step S41, when the absolute value of the discrete first-order gradient change rate is less than the set micro-amplitude disturbance threshold in multiple consecutive sampling control cycle values, it is determined that the fluctuation of the real-time power flow at the grid connection point is in the low-frequency noise range of active power flow. Step S42: Set the calibration value of the adaptive adjustment coefficient to 0, keep the active power compensation adjustment amount at 0, and control the photovoltaic inverter cluster to maintain the current output state.

3. The photovoltaic power generation prediction method based on multi-source data fusion according to claim 1, characterized in that, Step S4 also includes the following sub-steps: Step S43, when the risk coefficient of the power flow section exceeding the limit is greater than 0.90 and is within the second control range, shorten the sampling control cycle value; Step S44, set the sampling control cycle value to 0.2 times the standard reference cycle value, and update the active power compensation adjustment amount according to the shortened sampling control cycle value.

4. The photovoltaic power generation prediction method based on multi-source data fusion according to claim 1, characterized in that, It also includes the following steps: Step S5: Obtain the real-time frequency value of the grid connection point. When the real-time frequency value exceeds the set frequency limit threshold and the risk coefficient of the power flow section of the entire network exceeds the set safety threshold, stop adjusting the sampling control cycle value and adjust the active power output status of the photovoltaic inverter cluster according to the set droop control characteristic curve.

5. The photovoltaic power generation prediction method based on multi-source data fusion according to claim 1, characterized in that, Step S1 includes the following sub-steps: Step S11, obtain the interconnection impedance data of each node in the controlled network frame, and extract the eigenvalues ​​of the node admittance matrix of the tie line cross section under different power flow directions. Step S12: Calculate the power flow sensitivity coefficient of each controlled node to the power fluctuation of the tie line section based on the eigenvalues ​​of the node admittance matrix, and combine the power flow sensitivity coefficients to generate the active power-section power flow sensitivity matrix of the grid nodes.

6. The photovoltaic power generation prediction method based on multi-source data fusion according to claim 1, characterized in that, Step S2 includes the following sub-steps: Step S21, calculate the singular value decomposition of the active power-section power flow sensitivity matrix of the network nodes to generate the active power scheduling weight vector of the controlled nodes. Step S22: The expected value sequence of active power prediction and the active power scheduling weight vector of controlled nodes are introduced into the power flow safety verification logic in parallel to calculate the power flow distribution state of the entire network nodes. Based on the degree of deviation between the power flow distribution state and the network safety margin, the over-limit risk coefficient of the power flow section of the entire network is calculated.

7. The photovoltaic power generation prediction method based on multi-source data fusion according to claim 1, characterized in that, Step S3 includes the following sub-steps: Step S31, set a fixed number of sampling control cycle values ​​within the sliding control window; Step S32, obtain the active power deviation in the current time period, calculate the discrete difference of the active power deviation in multiple consecutive sampling control cycle values, and generate a discrete first-order rate of change.

8. The photovoltaic power generation prediction method based on multi-source data fusion according to claim 1, characterized in that, Step S4 includes the following sub-steps: Step S45, mapping the active power compensation adjustment amount to the power allocation coefficient of each photovoltaic inverter node, and sending it to the corresponding photovoltaic inverter node through the active power scheduling channel. Step S46: Each photovoltaic inverter node adjusts its own output power according to the power allocation coefficient to change the active power output state of the photovoltaic inverter cluster.

9. The photovoltaic power generation prediction method based on multi-source data fusion according to claim 1, characterized in that, Before step S1, the following steps are also included: Step S0, acquiring historical meteorological observation data, historical power output data and real-time operating data, normalizing and aligning the acquired data with time synchronization to generate a sequence of expected active power values.

10. A photovoltaic power plant power generation prediction system based on multi-source data fusion, used to implement the photovoltaic power generation prediction method based on multi-source data fusion as described in claim 1, characterized in that, include: The data feature purification module is used to acquire historical meteorological observation data, historical power output data, and real-time operating data, normalize and time-synchronize the acquired data to generate a sequence of expected active power prediction values. The network topology analysis module is used to obtain the topological admittance characteristic parameters of the network nodes, extract the active power flow direction characteristics of the tie line sections, construct the active power flow sensitivity matrix of the network nodes and generate the active power scheduling weight vector of the controlled nodes based on the active power flow sensitivity matrix of the network nodes. The power flow boundary verification module is used to introduce the expected value sequence of active power prediction and the active power scheduling weight vector of controlled nodes into the power flow safety verification logic in parallel, and simultaneously calculate the power flow section safety boundary to establish the risk coefficient of the power flow section exceeding the limit for the whole network. The transient deviation calculation module is used to obtain the real-time power flow power of the grid-connected point within the current sampling period, calculate the active power deviation between the real-time power flow power of the grid-connected point and the expected value sequence of active power prediction, and calculate the discrete first-order rate of change of the active power deviation within the sliding control window. The adaptive hedging scheduling control module is used to determine the control interval and adjust the sampling control cycle value based on the absolute value of the over-limit risk coefficient and the discrete first-order rate of change of the entire network power flow section, calculate the active power compensation adjustment amount, and control the active power output status of the photovoltaic inverter cluster.

Images