Low-voltage photovoltaic grid-connected voltage bidirectional collaborative optimization method and device considering load fluctuation

Abstract

The invention discloses a low-voltage photovoltaic grid-connected voltage bidirectional collaborative optimization method and device considering load fluctuation, relates to the technical field of photovoltaic energy storage, and aims to solve the problem of voltage out-of-limit caused by load fluctuation and unstable photovoltaic output after distributed photovoltaic large-scale grid connection. The method comprises the following steps: acquiring multi-node load data of a low-voltage distribution network, constructing a space-time Markov model, and outputting a load state transition probability and a time-space coupling coefficient; combining the coefficient with parameters such as photovoltaic array irradiance and environment temperature to construct a segmented prediction model, and outputting photovoltaic prediction output; constructing a multi-objective optimization function by taking photovoltaic prediction output as a constraint and taking a load fluctuation state factor as a weight, and solving to obtain an inverter and distribution network reactive power device adjusting instruction; and instructions are executed, voltage feedback is collected, data are returned to update model parameters, working condition changes can be dynamically adapted, and the operation and maintenance efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic energy storage technology, specifically to a method and apparatus for bidirectional collaborative optimization of low-voltage photovoltaic grid-connected voltage taking into account load fluctuations. Background Technology

[0002] Driven by the "dual carbon" goals, distributed photovoltaic (PV) systems have become widely adopted in low-voltage distribution networks due to their clean and efficient advantages, becoming a core component of the energy structure transformation of low-voltage distribution networks. However, voltage stability control remains a key bottleneck restricting the reliable application of low-voltage PV grid-connected systems.

[0003] Currently, existing technologies for low-voltage photovoltaic grid-connected voltage regulation have been studied, but they still have many limitations and cannot meet the precise regulation requirements under complex operating conditions. First, existing technologies lack the ability to perceive and model load fluctuations. Most solutions only focus on single-point load data collection and ignore the spatiotemporal correlation of loads at multiple nodes in the low-voltage distribution network. Load fluctuations at different nodes can affect each other, and existing monitoring systems lack quantitative analysis of the state transition patterns of multi-node loads and the coupling relationships between nodes. This results in low accuracy in predicting dynamic changes in the load, which in turn leads to lag in voltage regulation and an inability to respond promptly to voltage fluctuations caused by sudden load changes.

[0004] Secondly, there is a disconnect between photovoltaic power output prediction and load fluctuations. Existing photovoltaic power output prediction models mostly rely on environmental parameters such as irradiance and ambient temperature, without considering the indirect impact of load fluctuations on photovoltaic grid-connected operation. This results in a large deviation between the prediction results and the actual grid-connected operation, leading to unreasonable setting of subsequent voltage regulation constraints and further exacerbating voltage control errors.

[0005] Third, the coordinated control capability is weak. Existing solutions mostly adopt single control methods, such as only using reactive power compensation through photovoltaic inverters or relying solely on fixed reactive power devices in the distribution network. Moreover, the optimization objectives are singular, ignoring core needs such as reducing distribution network losses and controlling control costs. At the same time, the control strategies are not dynamically adjusted in conjunction with the degree of load fluctuation. For example, when the load fluctuates severely, conventional control weights are still used, which cannot prioritize ensuring voltage stability, resulting in low control efficiency and high operating costs.

[0006] Fourth, the control models have poor adaptability. Existing technologies mostly use fixed control parameters and lack a closed-loop iterative mechanism. When load characteristics, photovoltaic installed capacity, and other operating conditions undergo long-term changes, the models cannot update core parameters in real time, leading to a gradual decrease in control accuracy over time and making it difficult to adapt to the dynamic changes in low-voltage distribution networks. Furthermore, existing equipment operation and maintenance rely on on-site manual operation, lacking remote monitoring and maintenance capabilities. In scenarios where distributed photovoltaic nodes are dispersed, operation and maintenance efficiency is low and costs are high, further restricting the large-scale application of low-voltage photovoltaic grid-connected systems.

[0007] In summary, how to accurately perceive the spatiotemporal characteristics of load fluctuations, realize the correlation prediction between photovoltaic output and load fluctuations, construct a multi-objective collaborative control mechanism, and improve the model adaptability and operation and maintenance convenience have become the technical problems that urgently need to be solved in the field of low-voltage photovoltaic grid-connected voltage control. Summary of the Invention

[0008] The purpose of this invention is to provide a method and apparatus for bidirectional collaborative optimization of low-voltage photovoltaic grid-connected voltage that takes into account load fluctuations. The method involves a load fluctuation spatiotemporal sensing module that collects load data from distribution network nodes and models and outputs state transition probabilities and spatiotemporal coupling coefficients. A photovoltaic load coupling prediction module combines this data with photovoltaic environmental parameters to predict photovoltaic active power output. A multi-source collaborative optimization module uses this predicted output and load state factors to construct a multi-objective optimization function to solve for adjustment commands. A closed-loop control execution module issues commands and collects voltage feedback values. A model iteration update module optimizes the sensing module model based on the feedback values, thereby solving the aforementioned problems.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A bidirectional collaborative optimization method for low-voltage photovoltaic grid-connected voltage considering load fluctuations, characterized by comprising the following steps:

[0011] S1: Collect load data of each node in the low-voltage distribution network, construct a spatiotemporal Markov model, and output the state transition probability and spatiotemporal coupling coefficient.

[0012] S2: Based on the state transition probability and spatiotemporal coupling coefficient of step S1, a piecewise prediction model is constructed in combination with environmental parameters to output the predicted photovoltaic power output.

[0013] S3: Using the photovoltaic power prediction output in step 2 as a constraint and the load fluctuation status in step 1 as a weight, construct a multi-objective optimization function, solve for the adjustment command, and execute it.

[0014] S4: Execute the adjustment command of step 3, collect voltage feedback data and send it back to step 1 to update the model parameters of step 1, and realize closed-loop optimization of the whole process.

[0015] Step S1 involves collecting load data from each node of the low-voltage distribution network, constructing a spatiotemporal Markov model to output state transition probabilities and spatiotemporal coupling coefficients. Specifically, this includes collecting real-time data from the low-voltage photovoltaic grid-connected point and associated nodes, and collecting data from each node... , Active load power at time t reactive load power Duration of load fluctuation and the spatial distance between every two nodes ,in ;

[0016] A spatiotemporal Markov state transition model for load fluctuations is constructed based on spatiotemporal decomposition theory, outputting... Time of the first Each node is based on load status. Transition to state probability The formula is as follows:

[0017] ;

[0018] In the formula, Let be the probability of the load of node m at time t transitioning from state i to state j; α represents the actual number of transitions from state i to state j at time t; α is a smoothing coefficient, ranging from 0.1 to 0.5. The difference in active load between states i and j at node m; The reactive load difference between states i and j at node m; Let be the standard deviation of load power fluctuation at time t of node m; The spatial distance between the m-th and n-th nodes; This is the distance attenuation coefficient; Let K be the total number of occurrences of state i at node m at time t; K is the total number of load states, taking a value of 4, and is calculated as follows: It is classified into stable, mild, moderate, and severe.

[0019] Step S1 also includes simultaneously calculating the first and Node load spatiotemporal coupling coefficient The formula is as follows:

[0020] ;

[0021] In the formula, is the spatiotemporal coupling coefficient of the load at nodes m / n at time t, with a value ranging from 0 to 1; Let be the covariance of the active load at nodes m / n; , The variance of the active load at nodes m / n; , The duration of load fluctuation at time t for nodes m / n; This represents the average duration of historical load fluctuations.

[0022] Step S2, based on the state transition probabilities and spatiotemporal coupling coefficients of step S1, constructs a piecewise prediction model in conjunction with environmental parameters, and outputs the predicted photovoltaic power output. Specifically, this includes obtaining the photovoltaic array's... irradiance at any time Ambient temperature and component temperature Combined with the output of step 1 and Construct a segmented photovoltaic output coupling prediction model The formula is as follows:

[0023] ;

[0024] In the formula, The photovoltaic power output is predicted at time t+Δt; Rated power under standard test conditions for photovoltaic arrays; Predict the irradiance at time t+Δt; For standard irradiance, the value is taken as follows: W / m 2 ; The power temperature coefficient of the module is -0.3 to -0.5% / ℃. Let t be the temperature of the photovoltaic module at time t+Δt; For standard temperature, take the value. ℃; The multi-node spatiotemporal coupling correction term output in step 1; λ(t) is the load photovoltaic coupling coefficient, which takes a value of 0~0.3 and is fitted by the least squares method using historical data; The active load fluctuation at node m from t to t+Δt; Let be the photovoltaic module conversion efficiency at time t; ε is the convective heat transfer coefficient, with a value of 10~15; ε is the emissivity of the module, with a value of 0.8~0.9; σ is the Stefan-Boltzmann constant, with a value of 5.67×10. -8 ; The temperature of the sky.

[0025] Segmented photovoltaic output coupling prediction model in step S2 Medium component temperature The formula is derived from ambient temperature and irradiance:

[0026] ;

[0027] In the formula, Let t be the temperature of the photovoltaic module at time t+Δt; Predict the irradiance at time t+Δt; Standard temperature; Let be the photovoltaic module conversion efficiency at time t; ε is the convective heat transfer coefficient, with a value of 10~15; ε is the emissivity of the module, with a value of 0.8~0.9; σ is the Stefan-Boltzmann constant, with a value of 5.67×10. -8 ; The temperature of the sky.

[0028] Step S3 uses the photovoltaic power forecast output from Step 2 as a constraint and the load fluctuation state from Step 1 as a weighting basis to construct a multi-objective optimization function, solve for the adjustment command, and execute it. Specifically, the objectives include minimizing the grid connection point voltage deviation, minimizing distribution network losses, and minimizing adjustment costs, based on the load fluctuation state defined in Step S1 and the derived load fluctuation state factor. The specific values ​​under different load fluctuation conditions are: stable → 0.8, slight → 1.0, moderate → 1.2, severe → 1.5, and combined with step S2. The multi-objective optimization function is constructed as follows, and its formula is:

[0029] ;

[0030] The constraints are as follows:

[0031] ;

[0032] ;

[0033] ;

[0034] ;

[0035] In the formula, F is a multi-objective optimization function; The load state factor is the output of step S1; , , The target weight, and U(t) is the grid connection point voltage at time t. The rated voltage of the low-voltage system is 0.38kV; T is the optimization period. Let m be the active load at time t. The predicted power output of the photovoltaic system associated with node m is allocated by step 2; The line resistance from node m to the grid connection point; Let t be the inverter regulation cost. Adjusting costs per unit For fixed costs; The cost of reactive power adjustment in the distribution network at time t. Adjust costs per unit; The value is 0.35kV. The allowable voltage range is 0.42kV. The inverter provides reactive power output; This refers to the rated capacity of the inverter. To provide power to the reactive power compensation device in the distribution network; Power supply for the distribution network; Total distribution network loss;

[0036] The NSGA-II algorithm is used to solve the optimization function, and the inverter reactive power regulation command is output. and instructions for reactive power distribution devices , as the input for execution.

[0037] Step S4 executes the adjustment command of step S3, collects voltage feedback data and sends it back to step S1 to update the model parameters of step S1. Specifically, this includes updating the output of step 3. and The control actions are executed by sending data from edge computing nodes to inverters and reactive power distribution devices; and by collecting real-time voltage feedback values ​​from grid connection points. Calculate voltage deviation ;like ,in Then , The current load data is sent back to step S1 to update the formula. and Iteratively optimize the spatiotemporal Markov model in step S1; if Then, steps S1 to S3 will be executed according to the original cycle.

[0038] A bidirectional collaborative optimization device for low-voltage photovoltaic grid-connected voltage considering load fluctuations, used to implement the method described in any one of claims 1 to 7, characterized in that it comprises:

[0039] Load fluctuation spatiotemporal perception module: It consists of multiple distributed intelligent monitoring terminals, which collect the active load power, reactive load power, load fluctuation duration and spatial distance between each node; it has a built-in calculation model for step S1, and outputs the state transition probability and spatiotemporal coupling coefficient.

[0040] Photovoltaic load coupling prediction module: Receives state transition probabilities and spatiotemporal coupling coefficients from the spatiotemporal sensing module, as well as irradiance and ambient temperature data from the photovoltaic array. It incorporates the calculation model from step S2 and outputs... ;

[0041] Multi-source collaborative optimization module: Utilizing a Xilinx Kintex-7 FPGA chip, it receives data from the coupling prediction module. The load state factor of the time and space sensing module is embedded into the calculation model of step 3, and the inverter reactive power adjustment command and the distribution network reactive power device command are output.

[0042] Closed-loop control execution module: It includes an edge computing node and an RS485 / LoRa communication unit, which sends the inverter reactive power adjustment command and the distribution network reactive power device command to the execution device, and at the same time collects the voltage feedback value of the grid connection point;

[0043] Model Iteration Update Module: Receives the grid connection point voltage feedback value from the closed-loop control module, calculates its voltage deviation, feeds it back to the load fluctuation spatiotemporal sensing module, updates the load power fluctuation standard deviation and load spatiotemporal coupling coefficient, and optimizes the spatiotemporal Markov model.

[0044] The load fluctuation spatiotemporal sensing module has a power sensor sampling frequency ≥2kHz, voltage measurement accuracy ≤0.1%, and GPS positioning error ≤10m, ensuring the accuracy of spatiotemporal data; the photovoltaic-load coupling prediction module has a built-in temperature compensation circuit to correct... The formula for correcting measurement error is as follows: , This refers to the internal ambient temperature of the module.

[0045] It also includes a remote monitoring and maintenance module, which is built on edge computing nodes. The module displays the working status of each module through a web-based visual interface and can remotely start or stop the optimization process after obtaining scheduling permissions.

[0046] The specific mechanism is as follows:

[0047] First, the load fluctuation spatiotemporal sensing module, serving as the data acquisition and foundation, consists of multiple distributed intelligent monitoring terminals. It collects key load information from various related nodes in the low-voltage distribution network in real time, including the active and reactive load power of each node, the duration of load fluctuations, and the spatial distance between different nodes. The module embeds the calculation model of method step S1, which, based on the collected load data, analyzes and outputs two key results: one is the "load state transition probability," which quantifies the load transition from its current state at a given node; the other is the "load spatiotemporal coupling coefficient," which reflects the degree of correlation between load fluctuations between different nodes, providing a spatiotemporal characteristic basis for subsequent photovoltaic power output prediction.

[0048] Next, the photovoltaic load coupling prediction module receives the output data from the preceding modules. On one hand, it receives the load state transition probability and spatiotemporal coupling coefficient provided by the load fluctuation spatiotemporal perception module; on the other hand, it collects real-time environmental data of the photovoltaic array, including irradiance and ambient temperature, and embeds it into the calculation model of method step S2. This module combines the load fluctuation characteristics with the environmental influencing factors of photovoltaics, and finally outputs a "predicted photovoltaic active power output" that fits the actual operating conditions, ensuring that subsequent optimization decisions can be based on the accurate prediction of photovoltaic power output.

[0049] Subsequently, the multi-source collaborative optimization module, acting as the decision-making core, receives the predicted active power output of the photovoltaic load coupling prediction module and simultaneously acquires the "load state factor" analyzed by the load fluctuation spatiotemporal perception module. Different values ​​are set according to the degree of load fluctuation, such as lower values ​​for stable states and higher values ​​for heavily fluctuating states, and this is embedded into the calculation model of step S3. The module aims to minimize grid connection voltage deviation, distribution network losses, and regulation costs. Combining constraints such as allowable voltage range and reactive power output limits, it solves the problem using a specific optimization algorithm, ultimately outputting two types of key regulation commands: one for reactive power regulation of the photovoltaic inverter and the other for regulation of the distribution network reactive power compensation device, achieving collaborative control decisions for both photovoltaic and distribution network devices.

[0050] Subsequently, the closed-loop control execution module is responsible for command delivery and data feedback. It includes edge computing nodes and RS485 / LoRa communication units: on the one hand, it accurately sends the adjustment commands output by the multi-source collaborative optimization module to the actuators such as photovoltaic inverters and power grid reactive power compensation devices through the communication unit, driving the actuators to adjust the reactive power output; on the other hand, it collects the actual voltage feedback value of the grid connection point in real time, providing real operating condition data for subsequent model iterations.

[0051] Finally, the model iteration update module enables the device's adaptive optimization capability. It receives grid-connected voltage feedback values ​​from the closed-loop control execution module and calculates the deviation between the actual voltage and the rated voltage. If the deviation exceeds a set threshold, the voltage deviation, voltage feedback value, and current load data are transmitted back to the load fluctuation spatiotemporal perception module to update the key parameters of the module's internal model, further optimizing the accuracy of the spatiotemporal Markov model. If the deviation is within the allowable range, the "perception-prediction-optimization-execution" process is repeated according to the original cycle. In addition, the device is equipped with a remote monitoring and maintenance module. A web-based visual interface is built based on edge computing nodes, which can display the real-time working status of each module. After obtaining scheduling permissions, staff can remotely start or stop the optimization process, improving the convenience of device operation and maintenance and ensuring the voltage stability of the low-voltage photovoltaic grid-connected system.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] 1. Compared with existing technologies that mostly monitor single-point load data and are difficult to accurately capture dynamic changes in load, this invention collects multi-node load data through a load fluctuation spatiotemporal sensing module and constructs a spatiotemporal Markov model. It outputs the load state transition probability and spatiotemporal coupling coefficient, which can comprehensively quantify the temporal trend and spatial correlation of load fluctuations, providing a more accurate load-side basis for subsequent regulation and effectively avoiding the problem of voltage regulation lag caused by load fluctuation prediction deviation.

[0054] 2. Existing technologies often predict photovoltaic power output in isolation without considering load fluctuation characteristics, which can easily lead to a disconnect between photovoltaic power output prediction and actual operating conditions. This invention constructs a segmented prediction model by combining load spatiotemporal characteristics and photovoltaic environmental parameters through a photovoltaic load coupling prediction module, which significantly improves the accuracy of photovoltaic active power output prediction, provides reliable photovoltaic-side constraints for multi-objective optimization decision-making, and reduces voltage fluctuations caused by photovoltaic power output prediction deviations.

[0055] 3. Existing technologies mostly target single reactive power devices and have a single optimization objective. This invention introduces load state factors through a multi-source collaborative optimization module, dynamically adjusts the weights according to the degree of load fluctuation, and coordinates the regulation of photovoltaic inverters and reactive power devices in the distribution network. At the same time, it takes into account multiple objectives such as voltage deviation, network loss, and regulation cost. While ensuring the stability of low-voltage photovoltaic grid-connected voltage, it effectively reduces the operating loss and regulation cost of the distribution network.

[0056] 4. Existing control models are mostly fixed modes and difficult to adapt to dynamic changes in operating conditions. This invention collects voltage feedback data through a closed-loop control execution module and updates the core parameters of the load fluctuation spatiotemporal sensing module in real time through a model iteration update module. This enables dynamic iterative optimization of the control model, continuously improves the voltage control accuracy under different operating conditions, and enhances the device's adaptability to dynamic changes in load and photovoltaics.

[0057] 5. Compared with the existing technology that relies on on-site operation and has low efficiency, this invention is equipped with a remote monitoring and maintenance module. It builds a web-based visual interface based on edge computing nodes, supports remote monitoring of the module's working status and start / stop optimization process, greatly improves the convenience of operation and maintenance and reduces the cost of manual operation and maintenance. Attached Figure Description

[0058] Figure 1 This is a flowchart of the bidirectional collaborative optimization method for low-voltage photovoltaic grid-connected voltage that takes into account load fluctuations, as described in this invention. Detailed Implementation

[0059] The technical solutions of the present invention will now be described in detail with reference to the accompanying drawings.

[0060] like Figure 1 As shown, step S1 involves collecting load data from each node of the low-voltage distribution network, constructing a spatiotemporal Markov model to output state transition probabilities and spatiotemporal coupling coefficients. Specifically, this includes collecting real-time data from the low-voltage photovoltaic grid connection point and associated nodes, and collecting data from each node. , Active load power at time t reactive load power Duration of load fluctuation and the spatial distance between every two nodes ,in ;

[0061] A spatiotemporal Markov state transition model for load fluctuations is constructed based on spatiotemporal decomposition theory, outputting... Time of the first Each node is based on load status. Transition to state probability The formula is as follows:

[0062] ;

[0063] In the formula, Let be the probability of the load of node m at time t transitioning from state i to state j; α represents the actual number of transitions from state i to state j at time t; α is a smoothing coefficient, ranging from 0.1 to 0.5. The difference in active load between states i and j at node m; The reactive load difference between states i and j at node m; Let be the standard deviation of load power fluctuation at time t of node m; The spatial distance between the m-th and n-th nodes; This is the distance attenuation coefficient; Let K be the total number of occurrences of state i at node m at time t; K is the total number of load states, taking a value of 4, and is calculated as follows: It is classified into stable, mild, moderate, and severe.

[0064] Step S1 also includes simultaneously calculating the first and Node load spatiotemporal coupling coefficient The formula is as follows:

[0065] ;

[0066] In the formula, is the spatiotemporal coupling coefficient of the load at nodes m / n at time t, with a value ranging from 0 to 1; Let be the covariance of the active load at nodes m / n; , The variance of the active load at nodes m / n; , The duration of load fluctuation at time t for nodes m / n; This represents the average duration of historical load fluctuations.

[0067] Step S2, based on the state transition probabilities and spatiotemporal coupling coefficients of step S1, constructs a piecewise prediction model in conjunction with environmental parameters, and outputs the predicted photovoltaic power output. Specifically, this includes obtaining the photovoltaic array's... irradiance at any time Ambient temperature and component temperature Combined with the output of step 1 and Construct a segmented photovoltaic output coupling prediction model The formula is as follows:

[0068] ;

[0069] In the formula, The photovoltaic power output is predicted at time t+Δt; Rated power under standard test conditions for photovoltaic arrays; Predict the irradiance at time t+Δt; For standard irradiance, the value is taken as follows: W / m 2 ; The power temperature coefficient of the module is -0.3 to -0.5% / ℃. Let t be the temperature of the photovoltaic module at time t+Δt; For standard temperature, take the value. ℃; The multi-node spatiotemporal coupling correction term output in step 1; λ(t) is the load photovoltaic coupling coefficient, which takes a value of 0~0.3 and is fitted by the least squares method using historical data; The active load fluctuation at node m from t to t+Δt; Let be the photovoltaic module conversion efficiency at time t; ε is the convective heat transfer coefficient, with a value of 10~15; ε is the emissivity of the module, with a value of 0.8~0.9; σ is the Stefan-Boltzmann constant, with a value of 5.67×10. -8 ; The temperature of the sky.

[0070] Segmented photovoltaic output coupling prediction model in step S2 Medium component temperature The formula is derived from ambient temperature and irradiance:

[0071] ;

[0072] In the formula, Let t be the temperature of the photovoltaic module at time t+Δt; Predict the irradiance at time t+Δt; Standard temperature; Let be the photovoltaic module conversion efficiency at time t; ε is the convective heat transfer coefficient, with a value of 10~15; ε is the emissivity of the module, with a value of 0.8~0.9; σ is the Stefan-Boltzmann constant, with a value of 5.67×10. -8 ; The temperature of the sky.

[0073] Step S3 uses the photovoltaic power forecast output from Step 2 as a constraint and the load fluctuation state from Step 1 as a weighting basis to construct a multi-objective optimization function, solve for the adjustment command, and execute it. Specifically, the objectives include minimizing the grid connection point voltage deviation, minimizing distribution network losses, and minimizing adjustment costs, based on the load fluctuation state defined in Step S1 and the derived load fluctuation state factor. The specific values ​​under different load fluctuation conditions are: stable → 0.8, slight → 1.0, moderate → 1.2, severe → 1.5, and combined with step S2. The multi-objective optimization function is constructed as follows, and its formula is:

[0074] ;

[0075] The constraints are as follows:

[0076] ;

[0077] ;

[0078] ;

[0079] ;

[0080] In the formula, F is a multi-objective optimization function; The load state factor is the output of step S1; , , The target weight, and U(t) is the grid connection point voltage at time t. The rated voltage of the low-voltage system is 0.38kV; T is the optimization period. Let m be the active load at time t. The predicted power output of the photovoltaic system associated with node m is allocated by step 2; The line resistance from node m to the grid connection point; Let t be the inverter regulation cost. Adjusting costs per unit For fixed costs; The cost of reactive power adjustment in the distribution network at time t. Adjust costs per unit; The value is 0.35kV. The allowable voltage range is 0.42kV. The inverter provides reactive power output; This refers to the rated capacity of the inverter. To provide power to the reactive power compensation device in the distribution network; Power supply for the distribution network; Total distribution network loss;

[0081] The NSGA-II algorithm is used to solve the optimization function, and the inverter reactive power regulation command is output. and instructions for reactive power distribution devices , as the input for execution.

[0082] Step S4 executes the adjustment command of step S3, collects voltage feedback data and sends it back to step S1 to update the model parameters of step S1. Specifically, this includes updating the output of step 3. and The control actions are executed by sending data from edge computing nodes to inverters and reactive power distribution devices; and by collecting real-time voltage feedback values ​​from grid connection points. Calculate voltage deviation ;like ,in Then , The current load data is sent back to step S1 to update the formula. and Iteratively optimize the spatiotemporal Markov model in step S1; if Then, steps S1 to S3 will be executed according to the original cycle.

[0083] A bidirectional collaborative optimization device for low-voltage photovoltaic grid-connected voltage considering load fluctuations, used to implement the method described in any one of claims 1 to 7, characterized in that it comprises:

[0084] Load fluctuation spatiotemporal perception module: It consists of multiple distributed intelligent monitoring terminals, which collect the active load power, reactive load power, load fluctuation duration and spatial distance between each node; it has a built-in calculation model for step S1, and outputs the state transition probability and spatiotemporal coupling coefficient.

[0085] Photovoltaic load coupling prediction module: Receives state transition probabilities and spatiotemporal coupling coefficients from the spatiotemporal sensing module, as well as irradiance and ambient temperature data from the photovoltaic array. It incorporates the calculation model from step S2 and outputs... ;

[0086] Multi-source collaborative optimization module: Utilizing a Xilinx Kintex-7 FPGA chip, it receives data from the coupling prediction module. The load state factor of the time and space sensing module is embedded into the calculation model of step 3, and the inverter reactive power adjustment command and the distribution network reactive power device command are output.

[0087] Closed-loop control execution module: It includes an edge computing node and an RS485 / LoRa communication unit, which sends the inverter reactive power adjustment command and the distribution network reactive power device command to the execution device, and at the same time collects the voltage feedback value of the grid connection point;

[0088] Model Iteration Update Module: Receives the grid connection point voltage feedback value from the closed-loop control module, calculates its voltage deviation, feeds it back to the load fluctuation spatiotemporal sensing module, updates the load power fluctuation standard deviation and load spatiotemporal coupling coefficient, and optimizes the spatiotemporal Markov model.

[0089] The load fluctuation spatiotemporal sensing module has a power sensor sampling frequency ≥2kHz, voltage measurement accuracy ≤0.1%, and GPS positioning error ≤10m, ensuring the accuracy of spatiotemporal data; the photovoltaic-load coupling prediction module has a built-in temperature compensation circuit to correct... The formula for correcting measurement error is as follows: , This refers to the internal ambient temperature of the module.

[0090] It also includes a remote monitoring and maintenance module, which is built on edge computing nodes. The module displays the working status of each module through a web-based visual interface and can remotely start or stop the optimization process after obtaining scheduling permissions.

[0091] The specific implementation uses a low-voltage distribution network in a residential area of ​​a town, including 5 power consumption nodes, 5 distributed photovoltaic inverters with a total installed capacity of 250kVA, and 1 10kV / 0.38kV distribution transformer as the application scenario. It deploys a load fluctuation spatiotemporal sensing module, 5 DY-3000 intelligent monitoring terminals with a sampling frequency of 2kHz, a positioning error ≤10m, and a built-in spatiotemporal Markov model, and preset... , The photovoltaic load coupling prediction module collects multi-node load data and outputs state transition probabilities and spatiotemporal coupling coefficients. It combines environmental parameters and sensing module data to preset... Output photovoltaic predicted active power output, multi-source collaborative optimization module, embedded improved NSGA-II algorithm, preset and , , The system calculates the adjustment commands for the inverter and the reactive power distribution device. The closed-loop control execution module issues commands and collects voltage feedback values. The model iterative update module presets the parameters. When the voltage deviation exceeds the limit, the parameters of the sensing module are updated. At the same time, a remote monitoring and maintenance module is deployed to support web-based visualization and remote operation. During implementation, load spatiotemporal modeling, photovoltaic coupling prediction, multi-objective optimization, and closed-loop iteration are completed according to methods S1~S4. After one month of operation, the grid connection point voltage qualification rate increased from 82% to 98.5%, the average daily grid loss decreased by 12%, the average daily regulation cost decreased by 8%, and the number of on-site maintenance operations decreased by 60%, verifying the effectiveness of the solution.

Claims

1. A low-voltage photovoltaic grid-connected voltage bidirectional collaborative optimization method considering load fluctuation, characterized in that, Comprise the following steps: S1: Collecting load data of each node in low-voltage distribution network, constructing space-time Markov model output state transition probability and space-time coupling coefficient; S2: Based on the state transition probability and space-time coupling coefficient of step S1, combined with environmental parameters, a segmented prediction model is constructed to output photovoltaic predicted output; S3: With the photovoltaic predicted output of step 2 as the constraint, the load fluctuation state of step 1 as the weight, a multi-objective optimization function is constructed, and the adjustment instruction is solved and executed; S4: Execute the adjustment instruction of step 3, collect voltage feedback data back to step 1 to update the model parameters of step 1, realize the whole process closed loop optimization.

2. The low-voltage photovoltaic grid-connected voltage bidirectional collaborative optimization method considering load fluctuation according to claim 1, characterized in that, Step S1 collects load data of each node of the low-voltage distribution network, constructs a space-time Markov model to output state transition probability and space-time coupling coefficient, specifically including collecting real-time data of a low-voltage photovoltaic grid-connected point and associated nodes, and collecting each node , At time t, active load power , reactive load power , load fluctuation duration , and spatial distance between each two nodes , wherein ; Based on the theory of space-time decomposition, a space-time Markov state transition model of load fluctuation is constructed, and the output is the probability that the th node transfers from the load state to the state , and the formula is as follows: ​ ； In the formula, is the transition probability of the mth node load from state i to j at time t; is the actual transition number of the mth node i→j state at time t; α is the smoothing coefficient, taking values 0.1~0.5; is the active load difference of the mth node i / j state; is the reactive load difference of the mth node i / j state; is the load power fluctuation standard deviation of the mth node at time t; is the spatial distance between the mth and n th nodes; is the distance attenuation coefficient; is the total number of the mth node state i at time t; K is the total number of load states, taking values 4, and being divided into stable, mild, moderate, and severe according to ​ 3. The low-voltage photovoltaic grid-connected voltage bidirectional collaborative optimization method considering load fluctuation according to claim 2, characterized in that, Step S1 also includes simultaneously calculating the first With The spatio-temporal coupling coefficient of the load of the node The formula is as follows: ； In the formula, is the space-time coupling coefficient of m / n node load at time t, with a value of 0-1; is the covariance of m / n node active load; , is the variance of m / n node active load; , is the load fluctuation duration of m / n node at time t; is the average duration of historical load fluctuation.

4. The low-voltage photovoltaic grid-connected voltage bidirectional collaborative optimization method considering load fluctuation according to claim 1, characterized in that, Step S2 constructs a segmented prediction model based on the state transition probability and the space-time coupling coefficient of step S1, in combination with the environmental parameters, and outputs the photovoltaic prediction output. Specifically, the irradiance at the moment , the ambient temperature , and the component temperature are obtained, in combination with the and output by step 1, to construct a segmented photovoltaic output coupling prediction model , the formula of which is as follows: ； In the formula, is the predicted active power of the photovoltaic at time t+Δt; is the rated power of the photovoltaic array under standard test conditions; is the predicted irradiance at time t+Δt; is the standard irradiance, with a value of W / m²; is the power temperature coefficient of the component, with a value of -0.3~ -0.5% / ℃; is the temperature of the photovoltaic component at time t+Δt; is the standard temperature, with a value of ℃; is the multi-node space-time coupling correction term output by step 1; λ(t) is the load photovoltaic coupling coefficient, with a value of 0~0.3, fitted by the least square method of historical data; is the active load fluctuation of the mth node from t to t+Δt; is the conversion efficiency of the photovoltaic component at time t; is the convective heat transfer coefficient, with a value of 10~15; ε is the component emissivity, with a value of 0.8~0.9; σ is the Stefan-Boltzmann constant, with a value of 5.67×10 -8 ; is the sky temperature.

5. The low-voltage photovoltaic grid-connected voltage bidirectional collaborative optimization method considering load fluctuation according to claim 4, characterized in that, The segmented photovoltaic power output coupling prediction model of step S2 Medium assembly temperature Derived from ambient temperature and irradiance, the formula is: ； wherein, is the temperature of the photovoltaic module at time t+At; is the predicted irradiance at time t+At; is the standard temperature; is the conversion efficiency of the photovoltaic module at time t; is the convective heat transfer coefficient, taking a value of 10-15; ε is the emissivity of the module, taking a value of 0.8-0.9; σ is the Stefan-Boltzmann constant, taking a value of 5.67 x 10 -8 ; is the sky temperature.

6. The low-voltage photovoltaic grid-connected voltage bidirectional collaborative optimization method considering load fluctuation according to claim 1, characterized in that, Step S3, as a constraint of the photovoltaic predicted power of step 2 and a weight basis of the load fluctuation state of step 1, constructs a multi-objective optimization function, solves to obtain an adjustment instruction and executes, specifically including taking the minimum grid point voltage deviation, the minimum power distribution network loss and the lowest adjustment cost as the target, and according to the load fluctuation state divided in step S1 and the load fluctuation state factor obtained , the numerical values of which are specifically: smooth→0.8, mild→1.0, moderate→1.2, severe→1.5, and combining with the of step S2, the multi-objective optimization function is constructed as follows: ； The constraint conditions are as follows: ； ； ； ； wherein F is a multi-objective optimization function; is the load state factor output by step S1; , , is a target weight, and ; U(t) is the grid-connected point voltage at time t is a low-voltage system rated voltage, taking a value of 0.38 kV; T is an optimization period; is an active load of the mth node at time t is a predicted output of the associated photovoltaic of the mth node, distributed by step 2; is a line resistance from the mth node to the grid-connected point; is an inverter adjustment cost at time t is a unit adjustment cost is a fixed cost; is a distribution network reactive device adjustment cost at time t is a unit adjustment cost takes a value of 0.35 kV takes a value of 0.42 kV as a voltage allowable range; is an inverter reactive output; is an inverter rated capacity; is a distribution network reactive compensation device output; is a distribution network power supply power; is a distribution network total network loss; The NSGA-II algorithm is used to solve the optimization function, and the reactive power regulation instruction of the inverter is output and the instruction of the distribution network reactive device is taken as the execution input.

7. The low-voltage photovoltaic grid-connected voltage bidirectional collaborative optimization method considering load fluctuation according to claim 1, characterized in that, Step S4 executes the adjustment instruction of step S3, collects the voltage feedback data back to step S1 for updating the model parameters of step S1, specifically including the output of step 3 and The edge computing node is issued to the inverter and the distribution network reactive device, and the adjustment action is executed; the grid-connected point voltage feedback value is collected in real time , the voltage deviation is calculated; if , wherein , then , and the current load data are returned to step S1, and the and in the formula are updated, and the space-time Markov model of step S1 is iteratively optimized; if , steps S1-S3 are executed according to the original period.

8. A low-voltage photovoltaic grid-connected voltage two-way collaborative optimization device considering load fluctuation, used for realizing the method of any one of claims 1-7. Comprise: Load fluctuation space-time perception module: composed of multiple distributed intelligent monitoring terminals, collecting active load power, reactive load power, load fluctuation duration and spatial distance between each two nodes; Among them, the calculation model of step S1 is built in, and the state transition probability and space-time coupling coefficient are output; Photovoltaic load coupling prediction module: receive the state transition probability and the space-time coupling coefficient of the space-time perception module and the irradiance and environmental temperature data of the photovoltaic array, embed the calculation model of step S2, and output ; Multi-source collaborative optimization module: for the use of Xilinx Kintex-7 FPGA chip, receiving the load state factor of the coupling prediction module, embedding the calculation model of step 3, outputting the reactive power regulation instruction of the inverter and the instruction of the distribution network reactive power device; and the load state factor of the space-time perception module, embedding the calculation model of step 3, outputting the reactive power regulation instruction of the inverter and the instruction of the distribution network reactive power device; Closed loop control execution module: contains edge computing node and RS485 / LoRa communication unit, issues output inverter reactive power regulation instruction and distribution network reactive device instruction to the execution device, and collects grid-connected point voltage feedback value at the same time; Model iteration update module: receives the grid-connected point voltage feedback value of the closed loop control module, calculates the voltage deviation, feeds back to the load fluctuation space-time perception module, updates the load power fluctuation standard deviation and load space-time coupling coefficient, and optimizes the space-time Markov model. 9.The low-voltage photovoltaic grid-connected voltage bi-directional collaborative optimization device with load fluctuation considered according to claim 8, wherein, The power sensor sampling frequency of the load fluctuation space-time perception module is ≥2kHz, the voltage measurement accuracy is ≤0.1 level, the GPS positioning error is ≤10m, and the space-time data accuracy is ensured; the photovoltaic-load coupling prediction module is built-in temperature compensation circuit, which corrects measurement error, and the correction formula is , is the internal environment temperature of the module.

10. The low-voltage photovoltaic grid-connected voltage bidirectional cooperative optimization device with load fluctuation considered according to claim 8, characterized in that, It also includes a remote monitoring and operation module, which is constructed based on an edge computing node, and displays the working status of each module through a Web visual interface. After obtaining the dispatching authority, the optimization process can be remotely started or stopped.

Images