Multi-module parallel balancing and coordination control method, device, equipment and storage medium

Abstract

The present invention relates to a multi-module parallel equalization coordination control method, device, equipment and storage medium, including the following steps: performing time-series decoupling analysis on the multi-module parallel network based on the original state data to obtain a relative state feature set between modules; if the relative state feature set is not within their respective preset state ranges, then by using the Lagrange multiplier method, performing load balancing coordination analysis on the multi-module parallel network based on the relative state feature set to obtain a load balancing coordination allocation scheme; converting the load balancing coordination allocation scheme into control instructions to obtain a control instruction sequence for each module; parsing and implementing the control instruction sequence into each module to achieve load balancing of the multi-module parallel network, and solving the technical problem that when the grid conditions fluctuate or the load suddenly changes, the traditional method may not be able to respond in time, resulting in the aggravation of the uneven load distribution situation.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-module parallel networks, and particularly to a multi-module parallel balancing and coordination control method, device, equipment and storage medium. Background Art

[0002] Multi-module parallel networks play a crucial role in modern power electronic systems, especially in applications that require high power and high reliability. With the development of technology, more and more power electronic devices adopt a modular design, and multiple identical or different power modules are operated in parallel to achieve large-capacity output and redundant configuration. However, this architecture brings about the problem of load distribution: due to factors such as manufacturing tolerances, environmental changes, and aging effects, performance differences between modules are inevitable, which may cause some modules to bear excessive loads and fail prematurely, thus affecting the stability and reliability of the entire system. Therefore, how to ensure that these parallel modules can share the load evenly has become the focus of research.

[0003] Traditional load balancing methods often rely on static settings or simple feedback control, but such methods are difficult to adapt to complex dynamic working conditions and the requirements of instantaneous changes. In actual operation, when grid conditions fluctuate or the load suddenly changes, traditional methods may not be able to respond in a timely manner, resulting in an exacerbation of the uneven load distribution. In addition, the lack of in-depth understanding of the interaction between multiple modules also limits the effectiveness of existing technologies. This requires the development of a more intelligent and efficient load balancing and coordination control strategy that can monitor and adjust the working state of each module in real time to ensure a good load distribution even under non-ideal conditions, while improving the overall efficiency and lifespan of the system.

[0004] To solve the above problems, a new load balancing and coordination control method based on a multi-module parallel network is proposed. This method first uses a high-speed synchronous sampler to obtain the real-time operating parameters of each module in the network, and then performs time-series decoupling analysis to extract a dataset that reflects the relative state characteristics between modules. If it is found that these characteristics deviate from the preset safe range, the Lagrange multiplier method is introduced to perform load balancing and coordination analysis, formulate an optimized load distribution plan, and convert it into a specific control instruction sequence for application to each module. This method not only takes into account the dynamic interaction characteristics between different modules, but also provides a fast and accurate response mechanism, which is of great significance for improving the performance of multi-module parallel systems. Summary of the Invention

[0005] The main object of the present invention is to provide a multi-module parallel balancing and coordination control method, device, equipment and storage medium, so as to solve the technical problem that when grid conditions fluctuate or the load suddenly changes, traditional methods may not be able to respond in a timely manner, resulting in an exacerbation of the uneven load distribution.

[0006] To achieve the above object, the present invention provides a multi-module parallel equalization coordination control method, which is applied to a multi-module parallel network. The multi-module parallel network includes a plurality of interconnected modules, and the method includes the following steps:

[0007] Synchronously collect the operating parameters of the multi-module parallel network through a preset high-speed synchronous sampler to obtain original state data;

[0008] Perform time-series decoupling analysis on the multi-module parallel network based on the original state data to obtain a relative state feature set between the modules;

[0009] If the relative state feature set is not within their respective preset state ranges, then based on the relative state feature set, perform load balancing coordination analysis on the multi-module parallel network through the Lagrange multiplier method to obtain a load balancing coordination allocation scheme;

[0010] Convert the load balancing coordination allocation scheme into control instructions to obtain a control instruction sequence for each module;

[0011] Parse and implement the control instruction sequence into each module to achieve load balancing of the multi-module parallel network.

[0012] Further, the step of synchronously collecting the operating parameters of the multi-module parallel network through a preset high-speed synchronous sampler to obtain original state data includes:

[0013] Sample the voltage and current signals of each module through a high-speed synchronous sampler to obtain an original sampling sequence;

[0014] Perform phase compensation processing on the original sampling sequence based on a time synchronization trigger to obtain a synchronous sampling data stream;

[0015] Perform multi-scale decomposition on the synchronous sampling data stream through a wavelet packet decomposer to obtain multi-scale feature data;

[0016] Perform spectrum analysis on the multi-scale feature data based on the discrete Fourier transform to obtain original state data; wherein, the original state data includes fundamental wave parameters, harmonic contents, and phase sequence relationships of each module, and the phase sequence relationship is the phase sequence relationship between the three-phase voltage and the three-phase current inside each module.

[0017] Further, the step of performing time-series decoupling analysis on the multi-module parallel network based on the original state data to obtain a relative state feature set between the modules includes:

[0018] Perform a frequency-domain transformation on the original state data through a preset frequency-domain decomposer to obtain a frequency-domain feature vector, and perform principal component analysis on the frequency-domain feature vector to obtain a frequency-domain principal component set;

[0019] Perform a Hilbert transform on the frequency-domain principal component set to obtain an instantaneous frequency sequence;

[0020] Use Kalman filtering to perform state estimation on the instantaneous frequency sequence to obtain an estimated state quantity, and perform state space model analysis on the estimated state quantity to obtain a state transition matrix;

[0021] Based on the state transition matrix, perform time-series decoupling on the multi-module parallel network to obtain a decoupled state quantity, and perform covariance analysis on the decoupled state quantity to obtain the dynamic coupling relationship between modules;

[0022] Classify the dynamic coupling relationship between modules through fuzzy clustering analysis to obtain a clustering result, and perform inter-class similarity analysis on the clustering result to obtain a relative state feature set between modules; wherein, the relative state feature set includes power state features, electrical state characteristics, and dynamic state responses.

[0023] Further, through the Lagrange multiplier method, perform load balancing coordination analysis on the multi-module parallel network based on the relative state feature set to obtain a load balancing coordination allocation scheme, including:

[0024] Perform adaptive weight allocation on the relative state feature set to obtain a weighted feature matrix, and perform singular spectrum analysis on the weighted feature matrix to obtain a set of feature spectrum components;

[0025] Construct constraint conditions for the set of feature spectrum components through nonlinear programming to obtain an optimization objective function, and perform Lagrange multiplier expansion on the optimization objective function to obtain a set of Lagrangian functions, wherein the set of Lagrangian functions includes power balance constraints, voltage stability constraints, and frequency synchronization constraints;

[0026] Based on the Lagrange multiplier method, perform iterative solution on the set of Lagrangian functions to obtain an optimal solution sequence;

[0027] Based on the optimal solution sequence, perform load balancing coordination analysis on the multi-module parallel network to obtain a load balancing coordination allocation scheme; wherein, the load balancing coordination allocation scheme includes an optimal power allocation scheme, an optimal voltage regulation scheme, and an optimal frequency compensation scheme.

[0028] Further, the Lagrange multiplier expansion of the optimization objective function to obtain a set of Lagrangian functions includes:

[0029] Perform duality transformation on the optimized objective function to obtain a set of dual constraint functions, and perform gradient projection on the set of dual constraint functions to obtain a constraint gradient field;

[0030] Perform saddle point analysis on the constraint gradient field to obtain a sequence of saddle point solutions; wherein, the sequence of saddle point solutions includes a power saddle point solution, a voltage saddle point solution, and a frequency saddle point solution;

[0031] Perform Lagrangian expansion on the sequence of saddle point solutions based on the Wolfe duality theory to obtain a set of Lagrangian functions.

[0032] Further, the conversion of the load balancing coordination allocation scheme into control instructions to obtain a sequence of control instructions for each module includes:

[0033] Perform pulse width modulation conversion on the load balancing coordination allocation scheme through space vector decomposition to obtain a sequence of switching timings;

[0034] Perform dynamic compensation on the sequence of switching timings based on a state feedback matrix to obtain a compensation control quantity, and perform discrete quantization processing on the compensation control quantity to obtain a digital control word;

[0035] Perform feedforward prediction on the digital control word based on a preset predictive controller to obtain a predictive control sequence;

[0036] Perform parallel decoupling on the predictive control sequence through a module decoupler to obtain independent control quantities, and perform timing coordination on the independent control quantities to obtain a synchronous control sequence;

[0037] Perform hardware mapping on the synchronous control sequence based on an instruction generator to obtain a sequence of control instructions for each module.

[0038] Further, the parsing of the sequence of control instructions and the implementation into each module to achieve load balancing of the multi-module parallel network includes:

[0039] Perform real-time sharding processing on the sequence of control instructions to obtain a set of time sharding instructions, and perform parallel scheduling analysis on the set of time sharding instructions to obtain a module execution queue;

[0040] Perform priority sorting on the module execution queue to obtain a priority execution sequence, and perform deadlock detection and elimination on the priority execution sequence to obtain a deadlock-free execution sequence;

[0041] Perform instruction distribution on the deadlock-free execution sequence based on a preset distributed message bus to obtain module control messages, and generate a CRC check code for the module control messages to obtain a reliable transmission frame;

[0042] And perform state conversion on the reliable transmission frame to obtain an execution state sequence.

[0043] Implement the execution state sequence on each module to achieve load balancing of the multi-module parallel network.

[0044] The present invention also provides a multi-module parallel balancing and coordination control device, which is applied to a multi-module parallel network. The multi-module parallel network includes a plurality of interconnected modules, including:

[0045] An acquisition module, configured to synchronously acquire the operation parameters of the multi-module parallel network through a high-speed synchronous sampler to obtain original state data.

[0046] A first analysis module, configured to perform time-series decoupling analysis on the multi-module parallel network based on the original state data to obtain a relative state feature set between modules.

[0047] A second analysis module, configured to, if the relative state feature set is not within their respective preset state ranges, perform load balancing and coordination analysis on the multi-module parallel network based on the relative state feature set through the Lagrange multiplier method to obtain a load balancing and coordination allocation scheme.

[0048] A conversion module, configured to convert the load balancing and coordination allocation scheme into a control instruction sequence for each module.

[0049] An implementation module, configured to parse and implement the control instruction sequence on each module to achieve load balancing of the multi-module parallel network.

[0050] The present invention also provides a computer device, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the steps of the method described in any one of the above are implemented.

[0051] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in any one of the above are implemented.

[0052] The multi-module parallel balancing and coordination control method provided by the present invention includes the following steps: synchronously collecting the operation parameters of the multi-module parallel network through a high-speed synchronous sampler to obtain the original state data; performing time-series decoupling analysis on the multi-module parallel network based on the original state data to obtain the relative state feature set between the modules; if the relative state feature set is not within the respective preset state ranges, then through the Lagrange multiplier method, performing load balancing and coordination analysis on the multi-module parallel network based on the relative state feature set to obtain a load balancing and coordination distribution scheme; converting the load balancing and coordination distribution scheme into a control instruction sequence for each module; parsing and implementing the control instruction sequence into each module to achieve the load balancing of the multi-module parallel network, solving the technical problem that when the grid conditions fluctuate or the load changes suddenly, the traditional method may not be able to respond in time, resulting in the aggravation of the uneven load distribution situation, and realizing that by converting the load balancing and coordination distribution scheme into a specific control instruction sequence and parsing and implementing it into each module, this method realizes the refined management of the working state of each module, maximizes the potential of each module, improves the resource utilization rate, and at the same time reduces the maintenance cost and energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 is a schematic diagram of the steps of the multi-module parallel balancing and coordination control method in an embodiment of the present invention;

[0054] Figure 2 is a structural block diagram of the multi-module parallel balancing and coordination control device in an embodiment of the present invention;

[0055] Figure 3 is a schematic structural diagram of a computer device in an embodiment of the present invention.

[0056] The realization, functional characteristics and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0058] As Figure 1 shown, Figure 1 is a schematic diagram of the steps of a multi-module parallel balancing and coordination control method in an embodiment of the present invention;

[0059] In an embodiment of the present invention, a multi-module parallel equalization and coordination control method is provided, which is applied to a multi-module parallel network. The multi-module parallel network includes a plurality of modules connected to each other, and the method includes the following steps:

[0060] Step S1, synchronously collect the operation parameters of the multi-module parallel network through a preset high-speed synchronous sampler to obtain the original state data.

[0061] Specifically, in the process of implementing the multi-module parallel equalization and coordination control method, the key step of synchronously collecting the operation parameters of the multi-module parallel network through a high-speed synchronous sampler to obtain the original state data is crucial. In order to ensure that the load distribution among all modules can achieve an ideal equalization effect, it is first necessary to accurately understand the current working state of each module. The "high-speed synchronous sampler" mentioned here is a high-precision measurement tool specifically designed for power electronic systems. It can read the data of multiple points simultaneously within an extremely short time interval, ensuring that the data acquisition among different modules is strictly synchronous. For example, in a distributed power generation system composed of multi-module parallel inverters, these inverters may be connected to a common AC bus and deliver power to the power grid. Due to factors such as environmental temperature changes and load fluctuations, the actual output power of each inverter may vary. To maintain the stable operation of the system and optimize efficiency, we need to grasp the state of each inverter in real time. At this time, the high-speed synchronous sampler comes into play. It can sample the key parameters of each inverter, such as voltage and current, at the microsecond level or even faster. These parameters constitute the so-called "operation parameters". When the sampler completes the simultaneous acquisition of all relevant parameters, the obtained data set is the "original state data". These data not only contain the specific working conditions of each module at each moment, but also, because they are the result of synchronous acquisition, they are directly comparable to each other, which provides a solid foundation for subsequent time-series decoupling analysis. For example, if we find that the current value of a specific inverter abnormally increases during a certain period, then we can judge whether there is a problem and the root cause of the problem by comparing the data of other inverters at the same time point. In this way, based on the original state data, we can accurately identify which modules are in non-ideal working conditions, and then take measures to adjust their loads to achieve the load balance of the entire multi-module parallel network. Therefore, this step plays a fundamental role in the entire control method, ensuring that all subsequent operations can be based on a precise and synchronous data set.

[0062] Step S2, perform time-series decoupling analysis on the multi-module parallel network based on the original state data to obtain the relative state feature set between modules.

[0063] Specifically, after achieving the synchronous acquisition of the operating parameters of the multi-module parallel network and obtaining the original state data, the next crucial step is to perform a time-series decoupling analysis on the multi-module parallel network based on these original state data to obtain the relative state feature set between the modules. The core of this process lies in deeply analyzing the behavior patterns of each module over time and identifying their mutual influences, thereby providing a basis for subsequent load balancing coordination. For example, in a distributed power generation system composed of multi-module parallel inverters, we have obtained key parameters such as voltage and current of each inverter at different time points through a high-speed synchronous sampler. Now, we need to further understand the dynamic relationships behind these data. The time-series decoupling analysis is a method designed for this purpose, which decomposes the originally complex multi-variable time series into a form that is easier to process and understand. Specifically, it extracts the independent time evolution characteristics of each module from the original state data while revealing the potential connections between the modules. By deeply mining these data, we can find that some inverters may show consistent change trends under specific conditions, while others show different response patterns. Through this analysis, we can construct a data set reflecting the relative state features between the modules, namely the "relative state feature set". This feature set not only describes the state of a single module, but more importantly, it shows how the modules interact with each other and their differences. For example, when the grid conditions change, some inverters may quickly adjust their output power to adapt to the new demand, while others may respond slower or show different adjustment strategies. By comparing these relative state features, we can more accurately identify which modules are in the best working state, which need to be adjusted, and even predict potential problems that may occur in the future. Therefore, this analysis step lays the foundation for formulating an effective load balancing coordination plan, enabling us to flexibly adjust the working load of each module according to the actual working conditions to ensure the stability and efficient operation of the entire system. In summary, the time-series decoupling analysis based on the original state data is a complex but crucial link, which helps us clarify the operation details of the multi-module parallel network, extract valuable information, and thus guide subsequent optimization measures. This not only improves the overall performance of the system, but also provides strong support for realizing intelligent power management.

[0064] Step S3, if the relative state feature set is not within their respective preset state ranges, then through the Lagrange multiplier method, perform a load balancing coordination analysis on the multi-module parallel network based on the relative state feature set to obtain a load balancing coordination allocation plan.

[0065] Specifically, after completing the timing decoupling analysis of the multi-module parallel network and obtaining the relative state feature set among the modules, it is necessary to evaluate whether these features fall within the preset state range. If it is found that the relative state feature set is not within their respective preset state ranges, it means that the current load distribution may not achieve the ideal balanced state, and some modules may be overloaded or underloaded. At this time, in order to optimize the system performance, further measures need to be taken to adjust the working loads of each module. Specifically, by using the Lagrange multiplier method to conduct load balancing coordination analysis on the multi-module parallel network based on the relative state feature set is an important means to achieve this goal. The Lagrange multiplier method is a mathematical method for solving optimization problems with constraints, and here it is used to handle the complex and interrelated load distribution problems in the multi-module parallel network. This method can find an optimal load distribution scheme on the premise of keeping the total system output unchanged, making the working state of each module as close as possible to its optimal working point while meeting the overall constraints of the system, such as power balance, voltage stability, etc. For example, in a distributed power generation system composed of multi-module parallel inverters, assume that we have identified that the current values of some inverters exceed the normal range, which indicates that they are bearing too much or too little load. At this time, we can use the Lagrange multiplier method to construct an optimization model, taking the current load of each inverter as a variable and minimizing the load difference among all inverters as the objective function. At the same time, considering the requirements of the power grid and the safe operation requirements, we fix the total power output as a constant and ensure that the voltage level is maintained within an acceptable range. By solving this optimization problem, we can obtain a set of new load distribution suggestions, namely the "load balancing coordination distribution scheme". According to the obtained load balancing coordination distribution scheme, we can know which inverters need to increase or decrease their output power and the specific adjustment amplitude. In this way, even in the face of complex dynamic working condition changes, we can respond quickly, adjust the loads of each module in a timely manner, avoid the risk of failures caused by local overload, and improve the efficiency and stability of the entire system. Therefore, this step not only solves the problem of uneven existing load distribution but also provides a scientific basis for future continuous optimization, ensuring that the multi-module parallel network can always operate efficiently and reliably.

[0066] Step S4: Convert the load balancing coordination distribution scheme into control instructions to obtain the control instruction sequence of each module.

[0067] Specifically, after formulating the load balancing coordination allocation scheme, the next crucial step is to convert the load balancing coordination allocation scheme into control instructions to obtain the control instruction sequences for each module. This process is to ensure that the optimized load distribution strategy can be practically applied to the multi-module parallel network, enabling each module to adjust its working state in a predetermined manner, thereby achieving the ideal load balancing effect. Specifically, in a distributed power generation system composed of multi-module parallel inverters, we have obtained a scheme aimed at optimizing the load distribution of each inverter through the Lagrange multiplier method. However, this scheme exists in the form of theoretical values and cannot be directly used to control hardware devices. Therefore, we need to further process these values and convert them into specific control commands, namely "control instruction sequences". This step involves mapping each decision point in the load balancing coordination allocation scheme to specific operation instructions that the corresponding module can understand and execute. For example, if the optimization scheme indicates that a certain inverter needs to reduce its output power to avoid overload, the corresponding control instruction may be to adjust the PWM (pulse width modulation) duty cycle of the inverter or change its reference voltage setting value. To achieve this conversion, a set of predefined rules or algorithms are usually used, which can understand the results output by the optimization model and generate control signals suitable for a specific hardware platform accordingly. In this process, factors such as communication protocols, response times, and control accuracies between different inverters must be considered to ensure that all instructions can be accurately transmitted and executed. In addition, a mechanism needs to be designed to handle possible abnormal situations. For example, when a certain inverter fails to correctly receive or execute an instruction, the system should have the ability of self-diagnosis and repair to ensure the stable operation of the entire system. For example, in the above-mentioned distributed power generation system, assuming that after sequential decoupling analysis and optimization by the Lagrange multiplier method, we determine that the output power of some inverters needs to be adjusted. At this time, we convert these adjustment requirements into a series of precise control instructions, such as increasing or decreasing the current, adjusting the frequency, etc., and then send them to the corresponding inverters through an appropriate communication interface. After each inverter receives its own control instruction sequence, it will adjust its own operating parameters according to these instructions, thereby achieving the purpose of overall load balancing. In this way, not only an effective transition from theory to practice is realized, but also the entire multi-module parallel network can maintain an efficient and reliable working state under complex working conditions. In summary, the process of converting the load balancing coordination allocation scheme into control instructions is an important bridge connecting theory and practice. It ensures that the optimization results can effectively affect the actual operation of the system, enabling each module to adjust its behavior according to the optimal scheme, and ultimately achieving the best performance of the entire multi-module parallel network.

[0068] Step S5: Parse the control instruction sequence and implement it into each module to achieve load balancing of the multi-module parallel network.

[0069] Specifically, in the process of parsing and implementing the control instruction sequence into each module to achieve load balancing of the multi-module parallel network, the key lies in ensuring that each control instruction can be accurately understood and effectively executed. This process directly determines whether the optimized load distribution scheme can be correctly applied in the actual system, so as to achieve the expected performance improvement effect. Specifically, in a distributed power generation system composed of multi-module parallel inverters, we have obtained the specific control instruction sequence for each inverter through a series of steps. These instruction sequences include operations such as adjusting the output power and changing the operating frequency, but they exist in an abstract form and need to be further parsed to become actual signals that the inverter can recognize and respond to. To complete this task, a specially designed parsing unit or software module is usually adopted, which is responsible for interpreting the control instructions transformed from the load balancing coordination and distribution scheme and converting them into a format that conforms to the communication protocol of each inverter. For example, if a certain inverter uses the Modbus RTU communication protocol, the parsing unit will convert the general control instructions into a series of command frames under this protocol so that the inverter can correctly receive and process them. Once the parsing is completed, the next is the implementation stage. This not only involves sending the parsed instructions to the corresponding inverter, but also requires monitoring the response of each inverter to the instructions to ensure that they have adjusted their working states as expected. In this process, various challenges may be encountered, such as communication delays, data transmission errors, or hardware failures of individual inverters. To address these issues, the system should have a real-time feedback mechanism and self-correction ability. For example, when a certain inverter fails to correctly respond to the control instructions it receives, the system can automatically resend the instructions or take other remedial measures to ensure the stability and reliability of the entire system. For example, in the above-mentioned distributed power generation system, assume that we have generated a set of control instruction sequences according to the optimization results, instructing some inverters to increase the output power while others need to decrease it. The parsing unit will first convert these instructions into data packets suitable for the communication protocol of each inverter, and then send these data packets through a dedicated communication line. After each inverter receives its own control instructions, it will immediately start adjusting its working parameters, such as the PWM duty cycle or the reference voltage setting value, to adapt to the new load requirements. At the same time, the system will continuously monitor the status of each inverter to ensure that they have all been appropriately adjusted according to the predetermined plan. If any abnormality is found, the system will take corrective measures in a timely manner to ensure that all inverters can finally achieve the ideal load balancing state. To sum up, parsing and implementing the control instruction sequence into each module is the last and crucial link in achieving load balancing of the multi-module parallel network.It not only ensures that the optimization strategy can take root in the actual environment but also provides the necessary safeguard mechanism, enabling the entire system to maintain efficient and stable operation even in the face of complex and changing working conditions. Through this process, we can truly transform the theoretically optimized results into actual performance improvements, providing strong support for the application and development of the multi-module parallel system.

[0070] In a specific embodiment, the operation parameters of the multi-module parallel network are synchronously collected through a preset high-speed synchronous sampler to obtain the original state data, including:

[0071] Sampling the voltage and current signals of each module through a high-speed synchronous sampler to obtain an original sampling sequence;

[0072] Performing phase compensation processing on the original sampling sequence based on a time synchronization trigger to obtain a synchronous sampling data stream;

[0073] Performing multi-scale decomposition on the synchronous sampling data stream through a wavelet packet decomposer to obtain multi-scale feature data;

[0074] Performing spectrum analysis on the multi-scale feature data based on the discrete Fourier transform to obtain the original state data; wherein, the original state data includes the fundamental wave parameters, harmonic content, and phase sequence relationship of each module, and the phase sequence relationship is the phase sequence relationship between the three-phase voltage and the three-phase current inside each module.

[0075] Specifically, in a multi-module parallel network, to achieve efficient load balancing and coordinated control, it is first necessary to accurately and synchronously collect the operating parameters of each module. This process is completed by a high-speed synchronous sampler, which can not only quickly obtain the voltage and current signals of each module but also ensure the time consistency of these data, thus providing a solid foundation for subsequent analysis. Specifically, the process of synchronously collecting the operating parameters of the multi-module parallel network through the high-speed synchronous sampler to obtain the original state data includes multiple closely connected steps. First, the high-speed synchronous sampler samples the voltage and current signals of each module to obtain the original sampling sequence. At this stage, the sampler reads the three-phase voltage and three-phase current values at the output of each inverter simultaneously at an extremely high frequency (e.g., tens of thousands of times per second). This is because in a distributed power generation system composed of multiple multi-module parallel inverters, the operating state of each inverter may be affected by various factors, such as grid conditions, load changes, etc., and these effects will ultimately be reflected in the changes of voltage and current. Therefore, accurately capturing these dynamic changes is crucial for understanding the operating conditions of the entire system. The original sampling sequence obtained in this way contains a large amount of information about the real-time operating state of the inverter, but they have not been processed, and direct use may lead to difficulties in data analysis. Next, based on the time synchronization trigger, phase compensation processing is performed on the original sampling sequence to obtain the synchronous sampling data stream. Since there may be slight time differences between individual inverters, if these differences are not corrected, it will affect the subsequent data analysis results. The role of the time synchronization trigger is to eliminate this time deviation, so that data from different inverters can be compared on the same time basis. For example, in a distributed power generation system composed of multiple inverters, even if there are slight differences in the internal clocks of each inverter, through the processing of the time synchronization trigger, we can still ensure that the data of all inverters are synchronously collected, which provides a prerequisite for subsequent spectrum analysis and other forms of data mining. The synchronized sampling data stream not only retains all the information of the original sampling sequence but also enhances the consistency and comparability of data between different modules, which is particularly important for identifying and solving load imbalance problems. Subsequently, the synchronous sampling data stream is decomposed into multi-scale feature data through a wavelet packet decomposer. Wavelet packet decomposition is a powerful signal processing technology that can decompose complex electrical signals into multiple sub-bands according to different frequency ranges, thereby revealing various frequency components in the signal and their characteristics changing with time. In this example, the application of the wavelet packet decomposer means that we divide the originally continuous synchronous sampling data stream into a series of multi-scale feature data with different resolutions. Doing so can not only more carefully observe the transient behavior of each inverter but also more easily identify the useful information hidden under high-frequency noise.For example, when we focus on whether a certain inverter has abnormal vibration or overheating, multi-scale feature data can help us screen out specific frequency components related to these problems from a vast amount of raw data, thus providing a basis for fault diagnosis. Finally, spectrum analysis is performed on the multi-scale feature data based on the discrete Fourier transform to obtain the original state data. The discrete Fourier transform (DFT) is a classic algorithm for converting a time-domain signal into a frequency-domain representation. It can help us determine the fundamental wave parameters, harmonic content, and phase sequence relationship in the signal. Here, the "original state data" specifically refers to a data set containing the above key indicators, which reflect the interaction between the three-phase voltage and three-phase current within each module. Specifically, the fundamental wave parameters refer to the main frequency components of 50 Hz or 60 Hz (depending on the national grid standard); the harmonic content covers all integer multiples of the frequency higher than the fundamental wave frequency; the phase sequence relationship describes the relative angular difference between the three-phase voltage and three-phase current. By deeply analyzing these data, we can comprehensively understand the working state of each inverter, discover potential problems, and formulate reasonable optimization strategies accordingly. For example, in a distributed power generation system composed of multiple parallel-connected inverters, assume that we have completed all the above steps and obtained a complete set of original state data. These data show that there is an obvious imbalance in the three-phase current of one of the inverters, that is, the current of some phases is significantly higher than that of other phases. By further analyzing the fundamental wave parameters and phase sequence relationship of this inverter, we can confirm whether there is an internal fault or whether it is affected by external factors. If it is found that the temporary imbalance is caused by grid fluctuations, then the output of other inverters can be adjusted to compensate for this problem; but if it is due to a hardware failure of the inverter itself, timely maintenance measures need to be taken. In this way, we can make full use of the information provided by the original state data to ensure the stable operation and high efficiency of the entire system. To sum up, the process of obtaining the original state data by synchronously collecting the operating parameters of a multi-module parallel network through a high-speed synchronous sampler is a complex and delicate task. It involves knowledge and technologies in multiple professional fields, including high-precision sampling, time synchronization, multi-scale signal processing, and spectrum analysis. Each link plays an indispensable role and jointly constitutes a complete and effective data collection and analysis framework. This framework not only provides us with an opportunity to deeply understand the operation mechanism of the multi-module parallel system, but also lays a solid foundation for achieving accurate load balancing and coordinated control. In this way, we can ensure that the distributed power generation system is always in the best working state, meet the growing power demand, and at the same time provide strong technical support for the future construction of smart grids.

[0076] In a specific embodiment, the time-series decoupling analysis of the multi-module parallel network based on the original state data to obtain the relative state feature set between modules includes:

[0077] Perform a frequency-domain transformation on the original state data through a preset frequency-domain decomposer to obtain a frequency-domain feature vector, and perform principal component analysis on the frequency-domain feature vector to obtain a frequency-domain principal component set;

[0078] Perform a Hilbert transform on the frequency-domain principal component set to obtain an instantaneous frequency sequence;

[0079] Use Kalman filtering to perform state estimation on the instantaneous frequency sequence to obtain an estimated state quantity, and perform state-space model analysis on the estimated state quantity to obtain a state transition matrix;

[0080] Perform time-series decoupling on the multi-module parallel network based on the state transition matrix to obtain a decoupled state quantity, and perform covariance analysis on the decoupled state quantity to obtain the dynamic coupling relationship between modules;

[0081] Classify the dynamic coupling relationship between modules through fuzzy clustering analysis to obtain a clustering result, and perform inter-class similarity analysis on the clustering result to obtain the relative state feature set between modules; wherein, the relative state feature set includes power state features, electrical state characteristics, and dynamic state responses.

[0082] Specifically, to deeply understand the dynamic behavior and interaction among various modules in a multi-module parallel network, the process of performing time-series decoupling analysis on the multi-module parallel network based on the original state data is an essential step. This process can not only reveal the operating mechanism hidden behind complex power electronic systems but also provide a key basis for achieving efficient load balancing and coordinated control. Specifically, this step includes a series of closely related technical processes, from frequency-domain transformation to principal component analysis, then to Hilbert transform, Kalman filtering, state-space model analysis, and finally obtaining the relative state feature set among modules through fuzzy clustering analysis. First, the original state data is subjected to frequency-domain transformation by a frequency-domain decomposer to obtain a frequency-domain feature vector, and principal component analysis is performed on the frequency-domain feature vector to obtain a frequency-domain principal component set. At this stage, we use the frequency-domain decomposer to convert the time-series information in the original state data into a frequency-domain representation. For example, in a distributed generation system composed of multi-module parallel inverters, the three-phase voltage and current signals of each inverter contain rich frequency information, which reflects the power flow under different frequency components. By performing a fast Fourier transform (FFT) or discrete Fourier transform (DFT) on these signals, we can decompose them into a series of sine wave components of different frequencies, namely the so-called "frequency-domain feature vector". Subsequently, we apply principal component analysis (PCA) technology to screen out the main frequency components that can best represent the overall characteristics of the system and form a "frequency-domain principal component set". This step helps to reduce the amount of data for subsequent analysis while retaining the most important information, enabling us to focus more on the factors that have the greatest impact on system performance. Next, the frequency-domain principal component set is subjected to Hilbert transform to obtain an instantaneous frequency sequence. The Hilbert transform is a mathematical tool used to calculate the instantaneous amplitude and phase of a signal. Here, it is used to further analyze the time-evolution characteristics in the frequency-domain principal component set. Through the Hilbert transform, we can obtain the instantaneous frequency sequence of each main frequency component changing with time, which provides us with a more detailed understanding of the operating state of each inverter. For example, if the fundamental frequency of an inverter fluctuates, then this change will be directly reflected in its instantaneous frequency sequence. Such information is crucial for identifying the transient response and potential problems of the system because it can help us discover phenomena that are difficult to detect by static analysis alone. Then, Kalman filtering is used to perform state estimation on the instantaneous frequency sequence to obtain an estimated state quantity, and state-space model analysis is performed on the estimated state quantity to obtain a state transition matrix. Kalman filtering is a powerful recursive algorithm suitable for the optimal estimation of time-series data containing noise. In this process, we regard the instantaneous frequency sequence as part of the system output and assume that there is an underlying sequence of state variables behind it, which determines the changes in the actual observed values.Through Kalman filtering, we can extract the most likely state trajectory, i.e., the "estimated state quantity", from the observed data with random errors. Next, based on these estimated state quantities, we construct a state space model, which is a mathematical framework describing the internal dynamic behavior of the system. By analyzing the state transition matrix, we can understand how the various state variables interact with each other and their evolution laws over time. This is of great significance for predicting future system behavior and formulating corresponding control strategies. Immediately following, based on the state transition matrix, the multi-module parallel network is decoupled in time series to obtain the decoupled state quantities, and covariance analysis is performed on the decoupled state quantities to obtain the dynamic coupling relationship between the modules. The so-called "time series decoupling" means decomposing the originally complex multi-variable time series into several independent or weakly correlated sub-series for better understanding and processing. In this example, we reveal the internal connections between the inverters through the state transition matrix, and then use appropriate mathematical methods to separate these connections as much as possible to obtain a set of decoupled state quantities. After that, we perform covariance analysis on these decoupled state quantities to quantify the dynamic coupling strength between different modules. For example, if certain state quantities of two inverters show a high degree of correlation, then this may mean that there is some physical connection between them, such as sharing the same grid connection point or being affected by similar external disturbance sources. In this way, we can more clearly see how each module interacts, providing a basis for subsequent optimization and adjustment. Finally, the dynamic coupling relationship between the modules is classified through fuzzy clustering analysis to obtain the clustering results, and inter-class similarity analysis is performed on the clustering results to obtain the relative state feature set between the modules. Here, fuzzy clustering analysis is an unsupervised learning method that allows objects to belong to multiple categories and classifies them according to membership degrees. By performing fuzzy clustering on the dynamic coupling relationship between the modules, we can group the modules with similar behavior patterns into one category while retaining the subtle differences between them. Subsequently, we perform inter-class similarity analysis on these clustering results to find the common features and differences between different categories. The finally obtained "relative state feature set between the modules" covers multiple aspects such as power state characteristics, electrical state characteristics, and dynamic state responses, comprehensively reflecting the working conditions of each module and their mutual relationships. For example, in a distributed power generation system composed of multiple inverters, if we find that the power output of a certain type of inverter is relatively stable while another type shows greater fluctuations, then this may be due to their different working environments or task assignments. Through such analysis, we can take targeted measures to optimize the working conditions of each module to ensure the efficient operation of the entire system. In summary, the process of performing time series decoupling analysis on the multi-module parallel network based on the original state data is a multi-level and multi-dimensional data mining and modeling activity.It combines advanced technologies such as frequency-domain transformation, principal component analysis, Hilbert transform, Kalman filtering, state-space model analysis, and fuzzy clustering, aiming to deeply analyze the complex dynamic behaviors within the multi-module parallel system. Through this series of carefully designed steps, we can not only accurately capture the current working states of each module, but also predict their future development trends, thereby providing a scientific basis for achieving ideal load balancing and coordinated control. This method is not only applicable to the distributed generation systems mentioned above, but also provides valuable reference for other types of multi-module parallel systems, promoting the development of the field of intelligent power management and optimal control.

[0083] In a specific embodiment, by using the Lagrange multiplier method, based on the relative state feature set, load balancing and coordination analysis is performed on the multi-module parallel network to obtain a load balancing and coordination allocation scheme, including:

[0084] Performing adaptive weight allocation on the relative state feature set to obtain a weighted feature matrix, and performing singular spectrum analysis on the weighted feature matrix to obtain a set of feature spectrum components;

[0085] Constructing constraint conditions for the set of feature spectrum components through nonlinear programming to obtain an optimization objective function, and performing Lagrange multiplier expansion on the optimization objective function to obtain a set of Lagrange functions, including power balance constraint, voltage stability constraint, and frequency synchronization constraint;

[0086] Based on the Lagrange multiplier method, iteratively solving the set of Lagrange functions to obtain an optimal solution sequence, and based on the optimal solution sequence, performing load balancing and coordination analysis on the multi-module parallel network to obtain a load balancing and coordination allocation scheme; among them, the load balancing and coordination allocation scheme includes an optimal power allocation scheme, an optimal voltage regulation scheme, and an optimal frequency compensation scheme.

[0087] Specifically, to achieve load balancing and coordination in a multi-module parallel network, it is crucial to conduct an in-depth analysis of the relative state feature set through the Lagrange multiplier method. This process not only requires comprehensive consideration of the dynamic characteristics of each module but also ensures that the system reaches an optimal state in multiple aspects such as power balance, voltage stability, and frequency synchronization. Specifically, this step includes a series of complex technical processes from adaptive weight allocation to singular spectrum analysis, then to nonlinear programming, Lagrange multiplier expansion, iterative solution, and finally obtaining a load balancing and coordination allocation scheme. First, we perform adaptive weight allocation on the relative state feature set to obtain a weighted feature matrix and conduct singular spectrum analysis on the weighted feature matrix to obtain a set of eigen-spectrum components. At this stage, considering that there may be significant differences in the relative state features between different modules, we introduce an adaptive weight allocation mechanism. For example, in a distributed power generation system composed of multi-module parallel inverters, some inverters may exhibit different electrical characteristics and dynamic responses due to manufacturing tolerances or different working environments. By assigning corresponding weights to each feature, we can more accurately reflect these differences and thus construct a more reasonable "weighted feature matrix". Subsequently, using singular spectrum analysis (SSA), we decompose this matrix into multiple eigen-spectrum components, each of which represents a specific pattern or trend in the original data. This analysis method can help us identify which features have the greatest impact on the overall performance of the system and also provides a basis for further optimization. Next, through nonlinear programming, we construct constraint conditions for the set of eigen-spectrum components to obtain an optimization objective function and expand the optimization objective function using Lagrange multipliers to obtain a set of Lagrangian functions. What is involved here is to transform the above eigen-spectrum components into a form that can be used for mathematical modeling. Nonlinear programming is a powerful tool that allows us to find the optimal solution in the presence of multiple constraint conditions. In this example, we need to construct an optimization objective function that includes power balance constraints, voltage stability constraints, and frequency synchronization constraints. These constraint conditions reflect the basic requirements that a multi-module parallel system must meet: namely, the total output power remains constant, the voltage at each node is maintained within a safe range, and the operating frequencies of all modules are consistent. By expanding the optimization objective function using Lagrange multipliers, we transform it into a series of Lagrangian functions, which enables us to use the Lagrange multiplier method to solve the optimization problem with inequality constraints. For example, when we want a certain inverter to increase its output power, we must ensure that it does not cause overloading of other inverters or voltage fluctuations in the power grid at the same time. Such a multi-objective optimization problem can be effectively solved by the Lagrange multiplier method.Then, based on the Lagrange multiplier method, iterative solutions are obtained for the set of Lagrangian functions to get an optimal solution sequence, and based on the optimal solution sequence, load balancing coordination analysis is performed on the multi-module parallel network to obtain a load balancing coordination allocation scheme; wherein, the load balancing coordination allocation scheme includes an optimal power allocation scheme, an optimal voltage regulation scheme, and an optimal frequency compensation scheme. This process is one of approaching the optimal solution by repeatedly adjusting variable values. Since power electronic systems usually have highly nonlinear characteristics, it is very necessary to use iterative algorithms. Each iteration updates the parameters according to the gap between the current solution and the ideal solution until a solution that minimizes the objective function is found. In this process, we also need to continuously evaluate whether the obtained solution satisfies all the constraint conditions to verify the effectiveness of the algorithm. The finally obtained "optimal solution sequence" includes the optimal power allocation coefficient, voltage regulation coefficient, and frequency compensation coefficient, which jointly determine how each inverter should adjust its own operating state to achieve the best performance of the entire system. For example, if a higher power allocation coefficient is assigned to a certain inverter, it means that it should undertake more output tasks; while a lower voltage regulation coefficient indicates that the inverter does not need to respond too much to the grid voltage. Based on these optimal solutions, we further perform load balancing coordination analysis on the multi-module parallel network and finally form a complete "load balancing coordination allocation scheme". For example, in the distributed generation system mentioned above, assume that we have completed the adaptive weight allocation for the relative state feature set and obtained the feature spectrum component set through singular spectrum analysis. Then, we use nonlinear programming to construct an optimization objective function and perform Lagrange multiplier expansion to form a set of Lagrangian functions. Through iterative solutions, we find a set of optimal solution sequences, including an optimal power allocation scheme, an optimal voltage regulation scheme, and an optimal frequency compensation scheme. Specifically, if an inverter has a strong power output capacity and operates stably, it can be assigned a higher power output task; while for those inverters that are close to full-load operation, their output needs to be appropriately reduced to avoid overload risks. At the same time, we will also fine-tune the voltage and frequency settings of each inverter according to the actual needs of the power grid to ensure the stability and efficiency of the entire system. In this way, we can achieve the ideal load balancing effect and ensure that the multi-module parallel network is always in an efficient and stable operating state. To sum up, the process of performing load balancing coordination analysis on the relative state feature set by the Lagrange multiplier method is a highly complex optimization process that combines various advanced technologies such as adaptive weight allocation, singular spectrum analysis, nonlinear programming, Lagrange multiplier expansion, and iterative solutions. This process not only requires a solid mathematical foundation and technical support, but also requires us to flexibly apply these theoretical tools in practical applications to cope with various challenges existing in power electronic systems.Through this method, we can not only find the optimal solutions that satisfy all the constraint conditions, but also ensure that these solutions are optimal globally, providing a strong guarantee for achieving the load balancing coordination of the multi-module parallel network. This method is not only applicable to the distributed generation system mentioned above, but also provides valuable reference for other types of parallel systems of power electronic devices, promoting the development of the field of intelligent power management and optimal control.

[0088] In a specific embodiment, the Lagrangian multiplier expansion of the optimization objective function is performed to obtain a set of Lagrangian functions, including:

[0089] The duality transformation of the optimization objective function is carried out to obtain a set of dual constraint functions, and the gradient projection of the set of dual constraint functions is performed to obtain a constraint gradient field;

[0090] The saddle point analysis of the constraint gradient field is carried out to obtain a sequence of saddle point solutions; wherein, the sequence of saddle point solutions includes power saddle point solutions, voltage saddle point solutions and frequency saddle point solutions;

[0091] Based on the Wolfe duality theory, the Lagrangian expansion of the sequence of saddle point solutions is performed to obtain a set of Lagrangian functions.

[0092] Specifically, to deeply understand the process of performing Lagrangian multiplier expansion on the optimization objective function to obtain the set of Lagrangian functions, we need to explore in detail each step from duality transformation to gradient projection, then to saddle point analysis, and finally to complete the Lagrangian expansion based on the Wolfe duality theory. This process not only involves complex mathematical operations and the application of optimization theory but also provides a solid theoretical foundation for load balancing coordination in multi-module parallel networks. First, we perform a duality transformation on the optimization objective function to obtain a set of dual constraint functions, and then perform gradient projection on the set of dual constraint functions to obtain a constraint gradient field. At this stage, our goal is to transform the original optimization problem into a form that is easier to solve. Through the duality transformation, we can construct a new optimization problem, namely the "set of dual constraint functions", which is equivalent to the original problem in terms of the optimal solution but often has better mathematical properties. For example, in a distributed power generation system composed of multi-module parallel inverters, the original optimization objective function may contain multiple non-linear constraint conditions, such as power balance, voltage stability, and frequency synchronization. By introducing Lagrangian multipliers, we integrate these constraint conditions into the objective function to form an extended objective function. Subsequently, we use the duality principle to reformulate this extended objective function to obtain a new set of constraint conditions - the set of dual constraint functions. Next, to ensure that the obtained solution satisfies these constraint conditions, we perform gradient projection on the set of dual constraint functions. Gradient projection is a method used to keep the solution within the feasible region. It adjusts the variable values to minimize the objective function value while ensuring that all constraint conditions are satisfied. The result of this is the generation of a "constraint gradient field", which guides the direction of subsequent optimization algorithms. Then, we perform saddle point analysis on the constraint gradient field to obtain a sequence of saddle point solutions; among them, the sequence of saddle point solutions includes power saddle point solutions, voltage saddle point solutions, and frequency saddle point solutions. Saddle point analysis is an important means of identifying local extreme points in optimization problems, especially applicable to cases where there are multiple local optimal solutions. In this example, we are concerned with finding a solution that simultaneously satisfies all constraint conditions and minimizes the objective function. Due to the complexity and non-linear characteristics of power electronic systems, it is usually difficult to directly find such a global optimal solution. Therefore, we use the method of saddle point analysis to explore possible local optimal solutions. Specifically, by analyzing each point in the constraint gradient field, we can determine which points are both local minima and local maxima, and these points are called "saddle points". For a multi-module parallel inverter system, this means that we need to find a set of parameter settings such that the output power, grid voltage, and operating frequency of each inverter are within a reasonable range and the efficiency of the entire system is the highest. The resulting "sequence of saddle point solutions" corresponds to power saddle point solutions, voltage saddle point solutions, and frequency saddle point solutions respectively, which together form a potential set of optimal solutions.Then, based on the Wolfe duality theory, the saddle point solution sequence is Lagrangian expanded to obtain a set of Lagrangian functions. The Wolfe duality theory provides an effective framework for evaluating and improving the performance of optimization algorithms, especially when dealing with constrained nonlinear programming problems. According to this theory, we can further refine the saddle point solution sequence to ensure that the final solution obtained is not only locally optimal but also globally optimal. Specifically, we use the Lagrange multiplier method to expand the saddle point solution sequence. This step involves treating each saddle point solution as an optimal solution under specific conditions and considering the influence of other constraint conditions by introducing additional Lagrange multipliers. For example, if we have found a saddle point solution that satisfies power balance and voltage stability, then we need to check whether it also meets the requirements of frequency synchronization. In this way, we can construct a complete set of Lagrangian functions that comprehensively reflect the optimal solutions under all constraint conditions. The finally formed "set of Lagrangian functions" not only provides a clear goal for subsequent iterative solutions but also lays a solid theoretical foundation for achieving load balancing coordination in a multi-module parallel network. For example, in the distributed generation system mentioned above, assume that we have completed the duality transformation of the optimization objective function and obtained a set of dual constraint functions. By performing gradient projection on these constraint functions, we obtain the constraint gradient field, which indicates the optimization direction for us. Next, through saddle point analysis, we find a series of potential optimal solutions, including power saddle point solutions, voltage saddle point solutions, and frequency saddle point solutions. Finally, based on the Wolfe duality theory, we perform Lagrangian expansion on these saddle point solutions to obtain the final set of Lagrangian functions. In this process, we not only ensure that the operating state of each inverter is optimal but also verify the effectiveness of these states globally. For example, if an inverter performs best in terms of power output but we find that its voltage regulation or frequency compensation ability is insufficient, then through Lagrangian expansion, we can adjust the relevant parameters to achieve the best overall performance in the entire system. In this way, we can achieve the ideal load balancing effect and ensure the efficient and stable operation of the entire multi-module parallel network. In summary, the process of performing Lagrange multiplier expansion on the optimization objective function is a complex optimization process that combines duality transformation, gradient projection, saddle point analysis, and the application of the Wolfe duality theory. This process not only requires a solid mathematical foundation and technical support but also requires us to flexibly apply these theoretical tools in practical applications to address various challenges in power electronic systems. Through this method, we can not only find the optimal solutions that satisfy all constraint conditions but also ensure that these solutions are globally optimal, providing a strong guarantee for achieving load balancing coordination in a multi-module parallel network.This method is not only applicable to the distributed power generation system mentioned above, but also provides valuable reference for parallel systems of other types of power electronic devices, promoting the development of the field of intelligent power management and optimal control.

[0093] In a specific embodiment, the conversion of the load balancing coordination allocation scheme into control instructions to obtain the control instruction sequences of each module includes:

[0094] Performing pulse width modulation conversion on the load balancing coordination allocation scheme through space vector decomposition to obtain a switching timing sequence;

[0095] Performing dynamic compensation on the switching timing sequence based on a state feedback matrix to obtain a compensation control quantity, and performing discrete quantization processing on the compensation control quantity to obtain a digital control word;

[0096] Performing feedforward prediction on the digital control word based on a preset predictive controller to obtain a predictive control sequence;

[0097] Performing parallel decoupling on the predictive control sequence through a module decoupler to obtain independent control quantities, and performing timing coordination on the independent control quantities to obtain a synchronous control sequence;

[0098] Performing hardware mapping on the synchronous control sequence based on an instruction generator to obtain the control instruction sequences of each module.

[0099] Specifically, in order to convert the load balancing coordination allocation scheme into specific control instructions that can be understood and executed by each module, the entire process requires a series of fine-grained conversion and optimization steps. These steps include pulse width modulation (PWM) conversion through space vector decomposition, dynamic compensation based on the state feedback matrix, discrete quantization processing, feedforward prediction, parallel decoupling, and finally hardware mapping. Each link is closely connected, jointly ensuring that the optimized load distribution strategy can be implemented in the actual system. First, we perform PWM conversion on the load balancing coordination allocation scheme through space vector decomposition to obtain the switching timing sequence. At this stage, we need to convert the abstract load distribution scheme into specific operation instructions, that is, how to adjust the output power of each inverter. Space vector decomposition is a technology widely used in the field of power electronics. It can represent three-phase AC signals as rotating vectors on a two-dimensional plane, thus simplifying the design of PWM control. For example, in a distributed generation system composed of multiple parallel inverters, assuming that we have determined that a certain inverter needs to increase its output power. Then, through space vector decomposition, we can calculate the corresponding PWM duty cycle and generate a set of "switching timing sequences", which precisely specify the on-time and off-time of the inverter in each cycle. Such a control method can not only efficiently regulate the output power but also ensure the quality of the current waveform and reduce harmonic distortion. Next, based on the state feedback matrix, dynamic compensation is performed on the switching timing sequence to obtain the compensation control quantity, and discrete quantization processing is performed on the compensation control quantity to obtain the digital control word. Due to the dynamic characteristics of the power electronics system, simple PWM control may not be able to fully adapt to rapidly changing working conditions. Therefore, we introduce a state feedback mechanism to enhance control accuracy. The state feedback matrix can adjust control parameters in real time according to the current operating state of the system (such as voltage, current, etc.) to offset the influence brought by external disturbances or internal changes. For example, if the grid voltage suddenly fluctuates, the state feedback matrix will respond quickly and stabilize the output by modifying the PWM duty cycle. In addition, to make the control system compatible with digital hardware, we also need to perform discrete quantization processing on the compensation control quantity, that is, convert continuous analog signals into digital signals with a limited number of bits to form a "digital control word". This process ensures that all control instructions can be accurately executed by digital devices, and at the same time improves the anti-interference ability and stability of the system. Then, based on the predictive controller, feedforward prediction is performed on the digital control word to obtain the predictive control sequence. Predictive control is an advanced control strategy that uses a mathematical model to estimate the system behavior in the future for a period of time and adjusts the control input accordingly. For a multi-module parallel inverter system, this means that we can pre-calculate the best operations that each inverter should take in the next few cycles according to the current load demand and grid conditions.For example, if we anticipate an upcoming peak load period, the predictive controller can increase the output power of certain inverters in advance to avoid overload. In this way, we can not only improve the system's response speed but also better cope with the complex and changing working environment to ensure the stable operation of the entire system. Subsequently, the predictive control sequence is decoupled in parallel through a module decoupler to obtain independent control quantities, and the timing coordination of the independent control quantities is performed to obtain a synchronous control sequence. Since there may be mutual influences among the modules in a multi-module parallel system, directly applying the predictive control sequence may lead to unexpected behaviors. Therefore, we need to use a module decoupler to separate the originally associated control instructions into independent parts, namely "independent control quantities". This not only eliminates the coupling effect between modules but also enables each inverter to work according to its own optimal scheme. However, to ensure the synergy among all inverters, we also need to perform timing coordination on these independent control quantities to generate a set of "synchronous control sequences". For example, in a distributed generation system composed of multiple inverters, even if each inverter has its own control logic, we must ensure that their actions are synchronous to maintain the stability and efficiency of the entire system. Finally, based on the instruction generator, the synchronous control sequence is mapped to hardware to obtain the control instruction sequences of each module. The role of the instruction generator is to convert all the above optimization results into specific hardware commands, which can be directly sent to the control units of each inverter. For example, it may map each element in the synchronous control sequence to the value of a specific register or convert it into the data frame format under a communication protocol. In this way, after each inverter receives its own control instruction, it can immediately start adjusting its own operating parameters, such as the PWM duty cycle, reference voltage setting value, etc., to adapt to the new load requirements. Through this hardware mapping, we not only achieve an effective transition from theory to practice but also ensure that all control instructions can be accurately transmitted and executed. To sum up, the process of converting the load balancing coordination and distribution scheme into control instructions is a multi-level and multi-step technical implementation process. It combines various advanced technologies such as space vector decomposition, state feedback, discrete quantization, predictive control, module decoupling, and timing coordination, aiming to ensure that the optimized load distribution strategy can be accurately implemented in an actual power electronics system. This method is not only applicable to the above-mentioned distributed generation system but also provides valuable reference for other types of multi-module parallel systems, promoting the development of the field of intelligent power management and optimal control. In this way, we can ensure that the multi-module parallel network is always in an efficient and stable working state to meet the growing demands of modern power systems.

[0100] In a specific embodiment, parsing the control instruction sequence and implementing it into each module to achieve load balancing of the multi-module parallel network includes:

[0101] Performing real-time sharding processing on the control instruction sequence to obtain a time sharding instruction set, and performing parallel scheduling analysis on the time sharding instruction set to obtain a module execution queue;

[0102] Performing priority sorting on the module execution queue to obtain a priority execution sequence, and performing deadlock detection and elimination on the priority execution sequence to obtain a deadlock-free execution sequence;

[0103] Based on a preset distributed message bus, distributing the deadlock-free execution sequence to obtain module control messages, and generating a CRC check code for the module control messages to obtain a reliable transmission frame;

[0104] Performing state conversion on the reliable transmission frame to obtain an execution state sequence,

[0105] Implementing the execution state sequence into each module to achieve load balancing of the multi-module parallel network.

[0106] Specifically, in order to ensure that the control instruction sequence can be accurately parsed and effectively implemented in each module to achieve load balancing of a multi-module parallel network, the entire process needs to go through a series of closely connected steps such as real-time slicing processing, parallel scheduling analysis, priority sorting, deadlock detection and elimination, distributed message bus instruction distribution, CRC check code generation, and state transition. These steps together constitute an efficient and reliable control instruction execution framework, ensuring that the optimized load distribution strategy can be accurately applied in the actual system. First, we perform real-time slicing processing on the control instruction sequence to obtain a time-slicing instruction set, and perform parallel scheduling analysis on the time-slicing instruction set to obtain a module execution queue. At this stage, we need to divide the original continuous control instruction sequence into multiple smaller time segments, each of which contains a set of operation instructions that need to be executed within a specific time period. For example, in a distributed power generation system composed of multi-module parallel inverters, assume that we have generated a string of control instructions based on the load balancing coordination allocation scheme, and these instructions specify the specific operations of each inverter in the future. Through real-time slicing processing, we can divide this string of instructions into several "time-slicing instruction sets", each of which corresponds to a shorter time window. Next, in order to make full use of the parallel processing capabilities of the system, we perform parallel scheduling analysis on these time-sliced ​​instruction sets, evaluate the dependencies and execution order between each instruction, and finally build a set of "module execution queues". This not only improves the efficiency of instruction execution, but also enables the system to better adapt to rapidly changing working conditions. Then, the module execution queues are prioritized to obtain a priority execution sequence, and the priority execution sequence is deadlock detected and eliminated to obtain a deadlock-free execution sequence. Since there may be resource competition or task priority differences between different modules, simple sequential execution may not achieve the optimal effect. Therefore, we introduced a priority sorting mechanism to dynamically adjust the task priority of each module according to the current system operation status and task urgency. For example, if an inverter is processing a critical task, such as maintaining grid voltage stability, its priority may be higher than that of other inverters responsible for conventional power output. In addition, in order to avoid deadlock caused by improper task scheduling, we also need to detect and eliminate deadlock on the priority execution sequence. Deadlock refers to a state where two or more tasks wait for each other to release resources and fall into an infinite waiting state. By adopting preventive algorithms (such as the banker's algorithm) or detection and recovery mechanisms, we can identify potential deadlock situations and take measures to resolve them, thereby obtaining a set of "deadlock-free execution sequences". Then, based on the distributed message bus, the deadlock-free execution sequence is distributed with instructions to obtain module control messages, and CRC checksums are generated for the module control messages to obtain reliable transmission frames.A distributed message bus is a technology used to transfer information in a distributed system, which can provide efficient and reliable message communication services between different hardware components. In this example, we encapsulate each control instruction in the deadlock-free execution sequence into a "module control message" and send it to the corresponding inverter through the distributed message bus. To ensure that these instructions do not make errors or get lost during transmission, we also add a cyclic redundancy check (CRC) check code to each control message. The CRC check code is a widely used error detection method in the field of data transmission, which can effectively detect bit errors occurring during data transmission. Once the receiving end receives the control message with the CRC check code, it will recalculate the check value and compare it with the received check code. If the two are consistent, it means the data transmission is correct; otherwise, it indicates an error and the message needs to be resent. In this way, we generate a series of "reliable transmission frames", providing a solid guarantee for subsequent instruction execution. Subsequently, the state conversion of the said reliable transmission frames is performed to obtain an execution state sequence. State conversion means adjusting the current working state of the system according to the received control instructions to make it meet the expected goal. At this stage, after each inverter receives its own reliable transmission frame, it will update its operating parameters, such as the PWM duty cycle, reference voltage setting value, etc., according to the control instructions in it. At the same time, it will also record the changes in these parameters to form a set of "execution state sequences". For example, when an inverter receives an instruction to increase the output power, it will correspondingly adjust its PWM duty cycle and record this change for subsequent monitoring and feedback. The execution state sequence not only reflects the actual response of each inverter to the control instructions but also provides an important basis for the overall performance evaluation of the system. Finally, the said execution state sequence is implemented into each module to achieve the load balancing of the multi-module parallel network. Once all inverters have completed the state conversion and adjusted their working states according to the predetermined plan, the entire multi-module parallel network achieves an ideal load balancing effect. For example, in the above-mentioned distributed power generation system, through this series of carefully designed steps, we ensure that each inverter can work efficiently according to the optimized load distribution scheme, avoiding the risk of individual module overload and improving the operating efficiency and stability of the entire system. In this way, we have successfully transformed the abstract load balancing coordination and distribution scheme into specific control behaviors, achieving an effective transition from theory to practice. To sum up, parsing the control instruction sequence and implementing it into each module is a complex process involving multiple technical links. It combines various advanced technologies such as real-time sharding processing, parallel scheduling analysis, priority sorting, deadlock detection and elimination, distributed message bus instruction distribution, CRC check code generation, and state conversion, aiming to ensure that the optimized load distribution strategy can be accurately applied in the actual power electronics system.This method is not only applicable to the distributed power generation system mentioned above, but also provides valuable reference for other types of multi-module parallel systems, promoting the development of the field of intelligent power management and optimal control. In this way, we can ensure that the multi-module parallel network is always in an efficient and stable working state to meet the growing demands of modern power systems.

[0107] The above describes the multi-module parallel equalization and coordination control method in the embodiments of the present invention. Next, the multi-module parallel equalization and coordination control system in the embodiments of the present invention will be described. Please refer to Figure 2 , an embodiment of the multi-module parallel equalization and coordination control system in the embodiments of the present invention includes:

[0108] An acquisition module 21, configured to synchronously acquire the operation parameters of the multi-module parallel network through a preset high-speed synchronous sampler to obtain original state data;

[0109] A first analysis module 22, configured to perform time-series decoupling analysis on the multi-module parallel network based on the original state data to obtain a relative state feature set between modules;

[0110] A second analysis module 23, configured to, if the relative state feature set is not within their respective preset state ranges, perform load balancing and coordination analysis on the multi-module parallel network based on the relative state feature set through the Lagrange multiplier method to obtain a load balancing and coordination allocation scheme;

[0111] A conversion module 24, configured to convert the load balancing and coordination allocation scheme into a control instruction sequence for each module;

[0112] An implementation module 25, configured to parse and implement the control instruction sequence to each module to achieve load balancing of the multi-module parallel network.

[0113] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to that described in the above method embodiment, and details will not be repeated here.

[0114] Refer to Figure 3 , the embodiments of the present invention also provide a computer device, the internal structure of which can be as Figure 3As shown in the figure. The computer device includes a processor, a memory, a display screen, an input device, a network interface, and a database connected through a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program, when executed by the processor, implements the above method.

[0115] Those skilled in the art can understand that Figure 3 the structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0116] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0117] Those of ordinary skill in the art can understand that all or part of the processes in the above-described embodiment methods can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to memory, storage, database, or other media provided by the present invention and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0118] It should be noted that in this text, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article or method comprising a series of elements not only includes those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article or method. Without further limitation, an element defined by the phrase "comprising an..." does not exclude the presence of additional identical elements in the process, apparatus, article or method comprising such element.

[0119] The above are only the preferred embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. A multi-module parallel equalization and coordinated control method, characterized in that, Applied to a multi-module parallel network, the multi-module parallel network includes a plurality of interconnected modules, and comprises the following steps: Synchronously collect the operation parameters of the multi-module parallel network through a preset high-speed synchronous sampler to obtain original state data; Conduct time-series decoupling analysis on the multi-module parallel network based on the original state data to obtain a relative state feature set between the modules; If the relative state feature set is not within their respective preset state ranges, then through the Lagrange multiplier method, conduct load balancing coordination analysis on the multi-module parallel network based on the relative state feature set to obtain a load balancing coordination allocation scheme; Convert the load balancing coordination allocation scheme into control instructions to obtain a control instruction sequence for each module; Parse and implement the control instruction sequence into each module to achieve load balancing of the multi-module parallel network; The conducting time-series decoupling analysis on the multi-module parallel network based on the original state data to obtain a relative state feature set between the modules includes: Perform frequency-domain transformation on the original state data through a preset frequency-domain decomposer to obtain a frequency-domain feature vector, and conduct principal component analysis on the frequency-domain feature vector to obtain a frequency-domain principal component set; Perform Hilbert transform on the frequency-domain principal component set to obtain an instantaneous frequency sequence; Use Kalman filtering to perform state estimation on the instantaneous frequency sequence to obtain an estimated state quantity, and conduct state space model analysis on the estimated state quantity to obtain a state transition matrix; Based on the state transition matrix, conduct time-series decoupling on the multi-module parallel network to obtain a decoupled state quantity, and conduct covariance analysis on the decoupled state quantity to obtain a dynamic coupling relationship between the modules; Classify the dynamic coupling relationship between the modules through fuzzy clustering analysis to obtain a clustering result, and conduct inter-class similarity analysis on the clustering result to obtain a relative state feature set between the modules; wherein, the relative state feature set includes a power state feature, an electrical state characteristic, and a dynamic state response.

2. The multi-module parallel equalization coordination control method according to claim 1, characterized in that The synchronously collecting the operation parameters of the multi-module parallel network through a preset high-speed synchronous sampler to obtain original state data includes: Sample the voltage and current signals of each module through a high-speed synchronous sampler to obtain an original sampling sequence; Perform phase compensation processing on the original sampling sequence based on a time synchronization trigger to obtain a synchronous sampling data stream; Perform multi-scale decomposition on the synchronous sampling data stream through a wavelet packet decomposer to obtain multi-scale feature data; Perform spectrum analysis on the multi-scale feature data based on discrete Fourier transform to obtain original state data; wherein, the original state data includes fundamental wave parameters, harmonic contents, and phase sequence relationships of each module, and the phase sequence relationship is the phase sequence relationship between the three-phase voltage and the three-phase current inside each module.

3. The multi-module parallel equalization coordination control method according to claim 1, characterized in that The conducting load balancing coordination analysis on the multi-module parallel network through the Lagrange multiplier method based on the relative state feature set to obtain a load balancing coordination allocation scheme includes: Perform adaptive weight allocation on the relative state feature set to obtain a weighted feature matrix, and perform singular spectrum analysis on the weighted feature matrix to obtain a set of feature spectrum components; Construct constraint conditions for the set of feature spectrum components through nonlinear programming to obtain an optimization objective function, and perform Lagrangian multiplier expansion on the optimization objective function to obtain a set of Lagrangian functions, where the set of Lagrangian functions includes power balance constraints, voltage stability constraints, and frequency synchronization constraints; Based on the Lagrangian multiplier method, perform iterative solution on the set of Lagrangian functions to obtain an optimal solution sequence; Based on the optimal solution sequence, perform load balancing coordination analysis on the multi-module parallel network to obtain a load balancing coordination allocation scheme; wherein, the load balancing coordination allocation scheme includes an optimal power allocation scheme, an optimal voltage regulation scheme, and an optimal frequency compensation scheme.

4. The multi-module parallel equalization coordination control method according to claim 3, characterized in that The Lagrangian multiplier expansion of the optimization objective function to obtain a set of Lagrangian functions includes: Perform duality transformation on the optimization objective function to obtain a set of dual constraint functions, and perform gradient projection on the set of dual constraint functions to obtain a constraint gradient field; Perform saddle point analysis on the constraint gradient field to obtain a sequence of saddle point solutions; wherein, the sequence of saddle point solutions includes power saddle point solutions, voltage saddle point solutions, and frequency saddle point solutions; Based on the Wolfe duality theory, perform Lagrangian expansion on the sequence of saddle point solutions to obtain a set of Lagrangian functions.

5. The multi-module parallel equalization coordination control method according to claim 1, wherein The conversion of the load balancing coordination allocation scheme into control instructions to obtain a control instruction sequence for each module includes: Perform pulse width modulation conversion on the load balancing coordination allocation scheme through space vector decomposition to obtain a switching timing sequence; Based on a state feedback matrix, perform dynamic compensation on the switching timing sequence to obtain a compensation control quantity, and perform discrete quantization processing on the compensation control quantity to obtain a digital control word; Based on a preset predictive controller, perform feedforward prediction on the digital control word to obtain a predictive control sequence; Perform parallel decoupling on the predictive control sequence through a module decoupler to obtain independent control quantities, and perform timing coordination on the independent control quantities to obtain a synchronous control sequence; Based on an instruction generator, perform hardware mapping on the synchronous control sequence to obtain a control instruction sequence for each module.

6. The multi-module parallel equalization coordination control method according to claim 1, characterized in that The parsing of the control instruction sequence and its implementation into each module to achieve load balancing of the multi-module parallel network includes: Perform real-time sharding processing on the control instruction sequence to obtain a set of time sharding instructions, and perform parallel scheduling analysis on the set of time sharding instructions to obtain a module execution queue; Perform priority sorting on the module execution queue to obtain a priority execution sequence, and perform deadlock detection and elimination on the priority execution sequence to obtain a deadlock-free execution sequence; Based on a preset distributed message bus, distribute instructions for the deadlock-free execution sequence to obtain module control messages, and generate CRC check codes for the module control messages to obtain reliable transmission frames; And perform state conversion on the reliable transmission frame to obtain an execution state sequence. Implement the execution status sequence to each module to achieve load balancing of the multi-module parallel network.

7. A multi-module parallel equalization and coordination control device, characterized in that Applied to a multi-module parallel network, the multi-module parallel network includes a plurality of interconnected modules, including: An acquisition module, configured to synchronously acquire the operation parameters of the multi-module parallel network through a preset high-speed synchronous sampler to obtain original state data; A first analysis module, configured to perform time-series decoupling analysis on the multi-module parallel network based on the original state data to obtain a relative state feature set between the modules; A second analysis module, configured to, if the relative state feature set is not within their respective preset state ranges, perform load balancing coordination analysis on the multi-module parallel network based on the relative state feature set by using the Lagrange multiplier method to obtain a load balancing coordination allocation scheme; A conversion module, configured to convert the load balancing coordination allocation scheme into a control instruction sequence for each module; An implementation module, configured to parse and implement the control instruction sequence to each module to achieve load balancing of the multi-module parallel network; The performing time-series decoupling analysis on the multi-module parallel network based on the original state data to obtain a relative state feature set between the modules includes: Performing frequency-domain transformation on the original state data through a preset frequency-domain decomposer to obtain a frequency-domain feature vector, and performing principal component analysis on the frequency-domain feature vector to obtain a frequency-domain principal component set; Performing Hilbert transform on the frequency-domain principal component set to obtain an instantaneous frequency sequence; Performing state estimation on the instantaneous frequency sequence by using Kalman filtering to obtain an estimated state quantity, and performing state space model analysis on the estimated state quantity to obtain a state transition matrix; Performing time-series decoupling on the multi-module parallel network based on the state transition matrix to obtain a decoupled state quantity, and performing covariance analysis on the decoupled state quantity to obtain a dynamic coupling relationship between the modules; Classifying the dynamic coupling relationship between the modules through fuzzy clustering analysis to obtain a clustering result, and performing inter-class similarity analysis on the clustering result to obtain a relative state feature set between the modules; wherein, the relative state feature set includes a power state feature, an electrical state characteristic, and a dynamic state response.

8. A computer device, comprising a memory and a processor, wherein a computer program is stored in the memory, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images