Cooperative optimization control method, system and equipment for multi-mode power conversion device and storage medium

Abstract

The invention relates to the technical field of power conversion, in particular to a collaborative optimization control method, system and equipment of a multi-mode power conversion device and a storage medium. The method comprises the following steps: constructing a double-layer collaborative architecture by collecting multi-source heterogeneous operation parameters, obtaining a device operation state, power grid dynamic characteristics and environment change information, establishing a cross-level communication protocol to realize data synchronization, and outputting a real-time working condition feature vector; deploying a dynamic mode switching engine based on a real-time working condition feature vector, triggering multi-mode operation strategy reconstruction through a state transition probability model, constructing a self-adaptive switching mechanism comprising a parallel power supply mode, an island operation mode and a hybrid modulation mode, and establishing a composite fault response system to realize pre-diagnosis and grading disposal of an abnormal state; millisecond-level real-time control is implemented in the edge control layer to generate a pulse width modulation driving signal, multi-target global optimization is implemented in the cloud optimization layer to dynamically adjust an operation parameter combination, and closed-loop control is formed through bidirectional data interaction and strategy feedback.

Description

Technical Field

[0001] This invention relates to the field of power conversion technology, and in particular to a collaborative optimization control method, system, device, and storage medium for a multi-mode power conversion device. Background Technology

[0002] With the large-scale integration of renewable energy into the power system and the continuous increase in the penetration rate of fluctuating power sources such as photovoltaics and wind power, modern power grids are facing unprecedented operational challenges. Multimode power conversion devices, as key equipment connecting renewable energy to the power grid, need to achieve stable and reliable power conversion and transmission in complex and ever-changing grid environments. Traditional power conversion control methods mainly employ a single operating mode, which is difficult to adapt to rapid changes in grid operating conditions.

[0003] Current power conversion devices generally suffer from mode switching hysteresis. Under conditions of significant load fluctuations, traditional control strategies require a long time to complete mode switching, leading to a decrease in grid voltage and frequency stability. Furthermore, existing control methods lack a global optimization mechanism, resulting in ineffective coordination between devices and low overall operating efficiency. In terms of fault response, traditional protection strategies have long reaction times, making it difficult to quickly identify and handle complex fault states.

[0004] In existing technologies, the control strategies of power conversion devices are mainly based on local information, lacking the ability to perceive the overall operating status of the power grid. This limitation prevents the devices from adjusting their operating strategies according to the dynamic characteristics of the power grid, resulting in significant inadequacy in grid environments with a high proportion of renewable energy integration. Furthermore, traditional control architectures often employ centralized or fully distributed structures, making it difficult to simultaneously meet the dual requirements of real-time performance and global optimization. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention is proposed.

[0006] Therefore, the main technical problem addressed by this invention is to solve key issues in existing multimodal power conversion device control technologies, such as mode switching lag, lack of global collaborative optimization, and delayed fault response. By constructing an edge-cloud collaborative optimization architecture, intelligent adaptive switching of device operating modes is achieved, improving the operating efficiency and stability of power conversion devices in complex grid environments, and meeting the urgent need for flexible and reliable power conversion technology in high-proportion renewable energy grids.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] In a first aspect, embodiments of the present invention provide a collaborative optimization control method for a multi-mode power conversion device, which includes: collecting multi-source heterogeneous operating parameters and constructing a two-layer collaborative architecture; obtaining device operating status, grid dynamic characteristics and environmental change information; establishing a cross-level communication protocol to achieve data synchronization; and outputting a real-time operating condition feature vector.

[0009] Based on real-time operating condition feature vectors, a dynamic mode switching engine is deployed. Multi-modal operation strategy reconstruction is triggered by state transition probability model. An adaptive switching mechanism including parallel power supply mode, islanded operation mode and hybrid modulation mode is constructed. A composite fault response system is established to realize the pre-diagnosis and graded handling of abnormal states.

[0010] Millisecond-level real-time control is implemented at the edge control layer to generate pulse width modulation drive signals, and multi-objective global optimization is implemented at the cloud optimization layer to dynamically adjust the combination of operating parameters. Closed-loop control is formed through bidirectional data interaction and strategy feedback.

[0011] As a preferred embodiment of the collaborative optimization control method for the multi-mode power conversion device of the present invention, the multi-source heterogeneous operating parameters include device parameters, power grid parameters and environmental parameters. The device parameters cover modulation depth, switching frequency and junction temperature monitoring. The power grid parameters cover voltage imbalance, frequency deviation and load rate change. The environmental parameters cover wind speed and light intensity change.

[0012] The constructed dual-layer collaborative architecture includes an edge control layer and a cloud optimization layer. The edge control layer performs spatial vector pulse width modulation signal generation and instantaneous power calculation, while the cloud optimization layer generates a globally optimal set of weight coefficients through a multi-objective optimization algorithm.

[0013] As a preferred embodiment of the collaborative optimization control method for the multi-mode power conversion device described in this invention, the deployment of the dynamic mode switching engine based on real-time operating condition feature vectors includes constructing a multi-dimensional feature vector containing load rate, frequency deviation, harmonic distortion rate, power factor, junction temperature and state of charge based on a hidden Markov model, and triggering mode switching decisions through a state transition probability matrix.

[0014] Establish a three-level fault response mechanism: early warning level dynamic derating, protection level filter parameter reconfiguration, and emergency level bypass circuit switching.

[0015] As a preferred embodiment of the collaborative optimization control method for the multi-mode power conversion device of the present invention, the dynamic mode switching engine triggers a mode switching decision tree based on the multi-dimensional feature vector entering the critical region to control the power wave during the switching process.

[0016] The dynamic impedance matching algorithm is implemented by calculating the equivalent impedance in real time and adjusting the filter parameters, using a repetitive controller to compensate for harmonic components, and adjusting the switching frequency according to the junction temperature distribution.

[0017] As a preferred embodiment of the collaborative optimization control method for the multi-mode power conversion device described in this invention, the multi-mode operation strategy reconfiguration includes: in the parallel power supply mode, power balance distribution of multiple devices is achieved through a dynamic current sharing algorithm, and a redundancy mechanism is adopted to ensure high load conditions.

[0018] The islanded operation mode maintains grid voltage and frequency stability through a virtual synchronous machine model and autonomously switches operating states when the voltage at the point of common coupling is abnormal.

[0019] The hybrid modulation mode dynamically selects the optimal combination of equipment through a multi-dimensional energy efficiency assessment model, and adjusts the number of grid-connected devices and modulation depth according to changes in load rate.

[0020] As a preferred embodiment of the collaborative optimization control method for the multi-mode power conversion device described in this invention, the method involves establishing a composite fault response system by triggering a dynamic derating algorithm through junction temperature monitoring, reconstructing filter parameters through harmonic distortion rate detection, and triggering bypass circuit switching through DC bus voltage monitoring.

[0021] As a preferred embodiment of the collaborative optimization control method for the multimodal power conversion device described in this invention, the method further includes: an edge control layer performing space vector pulse width modulation drive signal generation, instantaneous power calculation, and overmodulation compensation algorithm at a preset period; and a cloud optimization layer periodically performing global multi-objective optimization and predicting operating trends by combining a digital twin model.

[0022] Secondly, embodiments of the present invention provide a collaborative optimization control system for a multi-mode power conversion device, which includes a multi-source data acquisition module, which acquires multi-source heterogeneous operating parameters and constructs a two-layer collaborative architecture, obtains device operating status, grid dynamic characteristics and environmental change information, establishes a cross-level communication protocol to achieve data synchronization, and outputs real-time operating condition feature vectors.

[0023] The module is built upon a dynamic mode switching engine deployed based on real-time operating condition feature vectors. It triggers multi-modal operation strategy reconstruction through a state transition probability model, constructs an adaptive switching mechanism that includes parallel power supply mode, islanded operation mode, and hybrid modulation mode, and establishes a composite fault response system to achieve pre-diagnosis and graded handling of abnormal states.

[0024] The parameter adjustment module performs millisecond-level real-time control at the edge control layer to generate pulse width modulation drive signals, and performs multi-objective global optimization at the cloud optimization layer to dynamically adjust the combination of operating parameters, forming a closed-loop control through bidirectional data interaction and strategy feedback.

[0025] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, they implement the steps of the cooperative optimization control method for a multi-mode power conversion device as described in the first aspect of the present invention.

[0026] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, they implement the steps of the cooperative optimization control method for a multi-mode power conversion device as described in the first aspect of the present invention.

[0027] The beneficial effects of this invention are as follows: This invention achieves real-time fusion of multi-source heterogeneous data by constructing a two-layer collaborative architecture, solving the problem of insufficient perception capability in traditional control methods; the dynamic mode switching engine based on a hidden Markov model realizes intelligent mode switching decision-making through a state transition probability matrix, solving the technical problem of mode switching lag in traditional methods and ensuring rapid adaptive response in complex power grid environments; the three-level fault response mechanism achieves early prediction and graded response to abnormal states through layered handling at the early warning, protection, and emergency levels; millisecond-level real-time control at the edge control layer ensures rapid response, global multi-objective optimization at the cloud optimization layer achieves optimal overall performance, and the closed-loop control formed by bidirectional data interaction addresses both real-time and global requirements. The overall solution solves the problem of collaborative control of power conversion devices in high-proportion renewable energy power grid environments. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 A flowchart of a collaborative optimization control method for a multi-mode power conversion device;

[0030] Figure 2 A computer device diagram illustrating the collaborative optimization control method for multimodal power conversion devices;

[0031] Figure 3 A topology diagram for a collaborative optimization control method for a multi-mode power conversion device;

[0032] Figure 4 Simulation diagram of converter operation mode switching for a collaborative optimization control method of multi-mode power conversion device. Detailed Implementation

[0033] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0034] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0035] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0036] Example 1

[0037] Reference Figure 1 - Figure 2 This is the first embodiment of the present invention, which provides a collaborative optimization control method for a multi-mode power conversion device, including:

[0038] S100: Collects multi-source heterogeneous operating parameters and constructs a two-layer collaborative architecture to obtain information on device operating status, power grid dynamic characteristics and environmental changes, establishes a cross-level communication protocol to achieve data synchronization, and outputs real-time operating condition feature vectors.

[0039] S200: Deploys a dynamic mode switching engine based on real-time operating condition feature vectors, triggers multi-modal operation strategy reconstruction through state transition probability model, constructs an adaptive switching mechanism including parallel power supply mode, islanded operation mode and hybrid modulation mode, and establishes a composite fault response system to realize the pre-diagnosis and graded handling of abnormal states.

[0040] S300: Millisecond-level real-time control is implemented at the edge control layer to generate pulse width modulation drive signals, and multi-objective global optimization is implemented at the cloud optimization layer to dynamically adjust the combination of operating parameters. Closed-loop control is formed through bidirectional data interaction and strategy feedback.

[0041] It should be noted that traditional multimode power conversion devices face significant challenges in high-proportion renewable energy grid environments, mainly in terms of delayed mode switching response, insufficient global collaborative optimization capabilities, and the lack of fault early warning mechanisms. Existing control strategies mostly adopt a single operating mode or a simple centralized architecture, which is difficult to adapt to the rapid changes in grid operating conditions and the complex operating conditions of large load rate fluctuations. When there are sudden load changes or grid faults, traditional control methods require a long time to complete mode switching, resulting in a decrease in voltage and frequency stability; at the same time, there is a lack of effective coordination mechanisms between devices, making it impossible to perform global optimization adjustments based on the overall dynamic characteristics of the grid. In complex fault conditions, the response time is long, making it difficult to achieve rapid identification and hierarchical handling.

[0042] Therefore, to address the aforementioned operational control and collaborative optimization issues, the S100-S300 steps construct an edge-cloud dual-layer collaborative architecture to achieve real-time fusion processing of multi-source heterogeneous data, acquiring comprehensive information on device operating status, grid dynamic characteristics, and environmental changes. A dynamic mode switching engine based on a Hidden Markov Model is deployed, triggering intelligent adaptive switching between parallel power supply, islanded operation, and hybrid modulation modes via a state transition probability matrix. A three-level fault response system is established to achieve pre-diagnosis and graded handling of abnormal states. Millisecond-level real-time control is implemented at the edge layer, and multi-objective global optimization is implemented at the cloud layer. Closed-loop control is formed through bidirectional data interaction and strategy feedback, effectively solving the shortcomings of traditional control methods in terms of real-time performance, globality, and reliability, and realizing intelligent collaborative optimization control of power conversion devices in complex grid environments.

[0043] Example 2

[0044] Reference Figure 1 - Figure 4 This is the second embodiment of the present invention.

[0045] In this embodiment, step S100, which involves collecting multi-source heterogeneous operating parameters and constructing a two-layer collaborative architecture, includes the following steps A1-A2:

[0046] A1: Multi-source heterogeneous operating parameters include device parameters, power grid parameters and environmental parameters. Device parameters cover modulation depth, switching frequency and junction temperature monitoring. Power grid parameters cover voltage imbalance, frequency deviation and load rate changes. Environmental parameters cover wind speed and light intensity changes.

[0047] Specifically, the multi-source heterogeneous data acquisition system achieves comprehensive parameter monitoring through distributed edge sensing terminals. Device parameter acquisition includes: modulation depth δ, with a real-time monitoring range of 0.8 ≤ δ ≤ 1.2, obtained through high-precision ADC sampling of PWM duty cycle changes; switching frequency f... swThe dynamic range is 2kHz-10kHz, with microsecond-level frequency measurement achieved through an FPGA built-in counter; the junction temperature (Tjunc) monitoring range is -40℃ to 125℃, verified by a combination of an infrared thermal imager and an embedded temperature sensor. Power grid parameter acquisition includes: three-phase voltage imbalance ε with a monitoring accuracy of ≤2%, achieving millisecond-level voltage amplitude and phase acquisition through a synchronous phasor measurement unit (PMU); frequency deviation Δf with a monitoring accuracy of ±0.2Hz, achieving high-precision frequency tracking using phase-locked loop (PLL) technology; and real-time load rate η with a monitoring range of 30% to 110%, acquired through a combination of smart meters and a power analyzer. Environmental parameter acquisition includes: wind speed v. w The monitoring range is 3m / s-25m / s, and an ultrasonic anemometer is used to achieve three-dimensional wind speed vector measurement; light intensity E lux Monitoring range 0-1000W / m 2 Spectral response characteristics are matched by using silicon photovoltaic cell arrays.

[0048] The data cleaning pipeline employs a multi-stage filtering architecture: the first-stage filter removes impulse noise using a median filter with a window length of 5 sampling points; the second-stage filter uses a Kalman filter to process Gaussian white noise, with the state equation X(k+1)=AX(k)+BU(k)+W(k) and the measurement equation Z(k)=HX(k)+V(k); the third-stage filter performs multi-scale noise reduction using wavelet transform, selecting the Daubechies-4 wavelet basis function, and setting the decomposition level to 6. The data synchronization mechanism adopts the IEEE 1588 precision clock protocol to achieve microsecond-level time synchronization accuracy, ensuring the timing consistency of multi-source data.

[0049] It should be noted that the multi-source heterogeneous parameter acquisition system handles the data fusion problem at different sampling frequencies through a time series alignment algorithm. The sampling frequency for device parameters is 10kHz, for power grid parameters it is 1kHz, and for environmental parameters it is 100Hz. These frequencies are unified to a 1kHz reference frequency through interpolation resampling technology. Data quality is assessed using both information entropy and signal-to-noise ratio (SNR). A data reacquisition mechanism is triggered when the information entropy H < 2.5 bits or the SNR < 20dB.

[0050] A2: The two-layer collaborative architecture includes an edge control layer and a cloud optimization layer. The edge control layer performs spatial vector pulse width modulation signal generation and instantaneous power calculation, while the cloud optimization layer generates a globally optimal set of weight coefficients through a multi-objective optimization algorithm.

[0051] Specifically, the dual-layer collaborative architecture achieves a balance between real-time performance and global capabilities through edge-cloud collaborative computing. The edge control layer deploys an ARM Cortex-A72 quad-core processor in conjunction with an FPGA coprocessor to generate Space Vector Pulse Width Modulation (SVPWM) signals with a 500μs control cycle. The SVPWM algorithm employs a seven-segment modulation strategy, generating three-phase modulated waves through sector determination and vector synthesis: first, the components Vα and Vβ of the reference voltage vector in the αβ coordinate system are calculated; then, the sector where the vector resides is determined, N = int(3θ / π), where θ = arctan(Vβ / Vα); finally, the basic voltage vector's action times T1 and T2 and the zero-vector time T0 are calculated. Instantaneous power calculation uses d... q Synchronous rotating coordinate system transformation:

[0052] P = 1.5(v) d ·i d +v q ·i q )

[0053] Q = 1.5(v) d ·i d +v q ·i q )

[0054] Three-phase to d-phase conversion is achieved through Park transformation. q Coordinate transformation of axes.

[0055] The cloud optimization layer deploys an Intel Xeon Platinum 8280 processor cluster, equipped with 256GB DDR4 memory and an NVMe SSD storage array, performing global multi-objective optimization every 5 minutes. An improved NSGA-II algorithm is used to solve the Pareto front.

[0056] min∑(α·P loss +β·THD 2 +γ·Δf 2 )

[0057] s.tV min ≤V i ≤V max

[0058] ∑P gen ≥1.05P load

[0059] An improved NSGA-II algorithm is used to solve the Pareto front and generate the optimal set of weight coefficients α, β, γ.

[0060] Algorithm parameter settings: population size 100, crossover probability 0.9, mutation probability 0.1, maximum number of iterations 200. The weight coefficient set {α,β,γ} is initially determined using the analytic hierarchy process (AHP) and then dynamically adjusted by the optimization algorithm.

[0061] Cross-layer communication employs a publish-subscribe model based on the MQTT protocol. The edge layer sends status data packets to the cloud every 100ms, and the cloud distributes optimization parameters every 5 minutes. The communication protocol uses JSON format encapsulation, achieving a data compression rate of 65% and keeping communication latency within 50ms. Fault tolerance mechanisms include reconnection after disconnection, data caching, and exception recovery to ensure the reliability of the communication link.

[0062] In an optional implementation, the two-layer collaborative architecture constructed in step S100 can also realize local autonomous decision-making capabilities through edge intelligent agent technology. When cloud communication is interrupted, the edge layer maintains basic control functions based on historical optimization parameters and local rule base to avoid system failure.

[0063] In another optional implementation, the multi-source heterogeneous operating parameter acquisition in step S100 can also be combined with an Internet of Things (IoT) sensor network to achieve remote environmental parameter monitoring through low-power wide area network technologies such as LoRaWAN or NB-IoT, which is particularly suitable for large-scale environmental data acquisition in distributed photovoltaic and wind power scenarios.

[0064] In this embodiment, step S200, which involves deploying a dynamic mode switching engine based on real-time operating condition feature vectors, includes the following steps B1-B4:

[0065] B1: The dynamic mode switching engine based on real-time operating condition feature vectors includes constructing a multi-dimensional feature vector containing load rate, frequency deviation, harmonic distortion rate, power factor, junction temperature and state of charge based on a hidden Markov model, and triggering mode switching decisions through a state transition probability matrix.

[0066] Specifically, the dynamic mode switching engine constructs a 6-dimensional feature vector based on a hidden Markov model:

[0067]

[0068] Define the state transition matrix:

[0069]

[0070] When the eigenvector enters the critical region |‖XX c When || < ε, the mode switching decision tree is triggered to ensure that the power fluctuation during the transition process is < 2%; the switching logic includes:

[0071] (1) Parallel → Islanding: PCC voltage dropped to 0.7 pu and did not recover within 200 ms.

[0072] (2) Island → Parallel: Synchronization phase difference Δθ < 5° and frequency deviation Δf < 0.2Hz for 10 seconds

[0073] (3) Hybrid → Parallel: Load rate prediction value η pred >90% for 30 seconds.

[0074] Establish a three-level fault response mechanism: early warning level dynamic derating, protection level filter parameter reconfiguration, and emergency level bypass circuit switching.

[0075] The composite fault response system includes a three-level handling mechanism:

[0076] Level 1 (Warning Level): When the junction temperature T junc Dynamic derating algorithm is activated when the temperature exceeds 80℃.

[0077]

[0078] Level 2 (Protection Level): When harmonic distortion rate (THD) > 5% is detected for 5 seconds, the reconstructed filter parameters are:

[0079] L new =L base ×(1+0.2·(THD-3))

[0080] C new =C base ×(1+0.15·(THD-3))Level3 (Emergency Level): DC bus voltage U occurs. dc When the value is >1.2, the thyristor bypass circuit is triggered and the standby device is switched within 2ms.

[0081] It should be noted that the state transition matrix of the Hidden Markov Model (HMM) is learned online through a sliding window mechanism with a window length of 1000 sample points. The transition probability is updated every 100 new samples. The model convergence criterion is ||pij(t+1)-pij(t)||F < 0.001, where ||〃||F represents the Frobenius norm. The dimensionality of the feature vector is reduced using Principal Component Analysis (PCA), retaining 95% of the cumulative variance contribution rate. In practical applications, a 6-dimensional feature vector can be compressed into 4 effective features.

[0082] Establish a three-level fault response mechanism: early warning level dynamic derating, protection level filter parameter reconfiguration, and emergency level bypass circuit switching.

[0083] Specifically, the three-level fault response mechanism is based on a risk assessment model to achieve graded triggering. Early warning level response: When the junction temperature T... juncThe dynamic derating algorithm is activated when the temperature exceeds 80℃. The derating factor is determined by the linear relationship between junction temperature and rated power; the power decreases by 5% for every 1℃ increase in temperature. Protection level response: When the harmonic distortion rate (THD) exceeds 5% for 5 seconds, the filter parameters are reconstructed, and the inductor parameter is adjusted to L. new =L base ×(1+0.2·(THD-3)), the capacitor parameter is adjusted to C. new =C base ×(1+0.15·(THD-3)), harmonic suppression is achieved through an adaptive filter. Emergency response: When the DC bus voltage U dc When the voltage is >1.2 pu, the thyristor bypass circuit is triggered within 2 ms, and the standby device switching sequence is started at the same time.

[0084] Fault diagnosis employs a fuzzy logic inference engine to establish a mapping relationship between fault symptoms and fault types. Input variables include four fuzzy sets: voltage deviation, current harmonics, temperature anomalies, and frequency fluctuations, described using triangular membership functions. The inference rule base contains 45 IF-THEN rules.

[0085] It should be noted that the trigger thresholds for the three-level fault response mechanism are determined based on historical fault statistics. The warning level threshold is set at 85% of the normal operating range, the protection level threshold at 75%, and the emergency level threshold at 60%. The response times are 100ms, 10ms, and 2ms, respectively, forming a progressive rapid response system. The fault prediction accuracy is evaluated using ROC curves, with an AUC value exceeding 0.92.

[0086] B2: Dynamic mode switching engine, which triggers the mode switching decision tree when a multi-dimensional feature vector enters the critical region, and controls the power wave during the switching process;

[0087] Data overlay logic:

[0088] F = 0.7·f trend (∑δ i )+0.3·f emergency (max(THD))

[0089] Among them, f trend Let f be the trend prediction function. emergency This is an emergency response function;

[0090] Scheduling strategy:

[0091] Under high load (η>90%), PQ control parallel mode is activated; under low load (η<30%), it switches to V / f control single mode; under normal operation, hybrid modulation mode is used, dynamically adjusting δ and f. sw combination.

[0092] Improved mode switching efficiency: shorter switching time, supporting 90% load fluctuation scenarios; enhanced global optimization capabilities: improved voltage qualification rate and reduced network loss; breakthrough in fault early warning: early warning of abnormal states based on hidden Markov models, reducing downtime due to faults.

[0093] The dynamic impedance matching algorithm is implemented by calculating the equivalent impedance in real time and adjusting the filter parameters, using a repetitive controller to compensate for harmonic components, and adjusting the switching frequency according to the junction temperature distribution.

[0094] The edge intelligent control unit implements device-level optimization, including:

[0095] Dynamic impedance matching algorithm: Real-time calculation of equivalent impedance: And adjust the L / C parameter to make |Z eq -Z grid |<0.1Ω

[0096] Harmonic suppression strategy: Use a repetitive controller to compensate for specific harmonics, with a gain coefficient of: hh represents the harmonic order;

[0097] Thermal equilibrium control: Adjusting the switching frequency based on the junction temperature distribution.

[0098] B3: Multimodal operation strategy reconfiguration includes parallel power supply mode to achieve power balance distribution of multiple devices through dynamic current sharing algorithm, and use redundancy mechanism to ensure high load conditions;

[0099] In parallel power supply mode, the N+1 redundancy mechanism is activated when the load rate η > 85%, and a dynamic current sharing algorithm is adopted:

[0100]

[0101] Z i Let P be the equivalent impedance of the i-th device. A current sharing accuracy of ±1% is achieved by adjusting the pulse width modulation ratio D (0.4 ≤ D ≤ 0.6). When the power deviation P... dev Power balance adjustment is initiated when the threshold is exceeded by 5%.

[0102] The islanded operation mode maintains grid voltage and frequency stability through a virtual synchronous machine model and autonomously switches operating states when the voltage at the point of common coupling is abnormal.

[0103] The islanded operation mode is triggered when the voltage at the point of common coupling drops to 0.8 pu, constructing a virtual synchronous machine model:

[0104]

[0105] The moment of inertia is set to J = 6.5 kg·m. 2The damping coefficient D = 12.5 N·m·s / rad ensures that the frequency fluctuation is ≤ ±0.5 Hz;

[0106] The hybrid modulation mode dynamically selects the optimal combination of equipment through a multi-dimensional energy efficiency assessment model, and adjusts the number of grid-connected devices and modulation depth according to changes in load rate.

[0107] The hybrid modulation mode is activated when the load rate is 30% < η < 85%, establishing a multi-dimensional energy efficiency evaluation model:

[0108]

[0109] When E opt If the value is greater than 0.95, the current device combination is maintained; otherwise, the topology reconfiguration algorithm is activated to dynamically adjust the number of grid-connected devices and the modulation depth δ (0.8≤δ≤1.2).

[0110] B4: Establish a composite fault response system, trigger a dynamic derating algorithm by monitoring junction temperature, reconstruct filter parameters by detecting harmonic distortion rate, and trigger bypass circuit switching by monitoring DC bus voltage.

[0111] Specifically, the composite fault response system employs multi-sensor information fusion technology and uses DS evidence theory to analyze the credibility of fault criteria. The basic probability assignment function m(A) represents the degree of confidence in fault type A, and the combination rule is as follows:

[0112]

[0113] Where K = ∑m1(B)·m2(C) is the conflict factor. Fault types include four categories: over-temperature faults, harmonic faults, over-voltage faults, and communication faults.

[0114] Junction temperature monitoring employs dual detection using an infrared thermal imager and an embedded thermistor, with a sampling frequency of 100Hz and a temperature accuracy of ±0.5℃. The dynamic derating algorithm utilizes piecewise linear control: when 75℃... <T junc At ≤85℃, When 85℃ <T junc When c≤95℃, P derate =P rated ×(0.8-0.4(T junc -75)); When T junc At >95℃, P derate =0.4×P rated The derating rate is controlled within 10 MW / min to avoid the impact of power surges on grid stability.

[0115] Harmonic distortion rate detection employs a sliding DFT algorithm with a window length set to one fundamental frequency period and an update frequency of 50Hz. Filter parameter reconstruction is based on the principle of an adaptive notch filter.

[0116] L new =L base ×(1+0.2·(THD-3))

[0117] C new =C base ×(1+0.15·(THD-3))

[0118] Where K1 = 0.2 and K2 = 0.15 are adjustment coefficients. The parameter update adopts the recursive least squares (RLS) algorithm: θ(k+1) = θ(k) + K(k)〃e(k), where K(k) is the gain matrix and e(k) is the prediction error.

[0119] DC bus voltage monitoring uses a Hall effect voltage sensor with a sampling frequency of 10kHz and a voltage accuracy of 0.1%. Bypass circuit switching uses a fast thyristor switch with a response time of <2ms. Switching logic: When U... dc A Level-1 warning is triggered when the CPU usage persists for 5ms. dc Bypass switching is triggered immediately when the power consumption is >1.3 PU. The standby device startup sequence is: pre-charging (50ms) – synchronization detection (100ms) – grid connection switching (50ms), with a total switching time <200ms. The fault warning system uses a Support Vector Machine (SVM) classifier to identify fault modes. The feature vector includes 12 features such as voltage amplitude, effective current value, power factor, temperature gradient, and harmonic content. The SVM kernel function uses a radial basis function (RBF).

[0120] K(X i ,X j )=exp(-γ·‖X i -X j || 2 )

[0121] Where γ = 0.1 is the kernel parameter. The training samples include three classes: normal state, minor fault, and severe fault, with a sample ratio of 6:3:1. The hyperparameter C = 100 and γ = 0.1 were optimized through grid search. The classification accuracy reached 96.5%, and the false alarm rate was <2%. It should be noted that the design of the composite fault response system follows three principles: speed, accuracy, and reliability. Speed ​​is achieved through hardware acceleration and algorithm optimization to achieve millisecond-level response; accuracy is improved through multi-source information fusion and machine learning methods to enhance fault identification accuracy; and reliability is ensured through redundant design and fault isolation mechanisms to ensure the safe operation of the system. The fault recovery strategy adopts a hierarchical recovery mode: equipment-level recovery, system-level recovery, and grid-level recovery, with recovery times in seconds, minutes, and hours, respectively.

[0122] In one optional implementation, the deployment of the dynamic mode switching engine based on real-time operating condition feature vectors in step S200 can also be combined with the deep reinforcement learning (DQN) algorithm to learn the optimal switching strategy through interaction between the agent and the environment, which is suitable for complex and ever-changing power grid operating environments.

[0123] In another optional implementation, the multimodal operation strategy reconstruction in step S200 can also be based on modeling the topological connections between devices using a graph neural network (GNN), and distributed cooperative control can be achieved through a message passing mechanism to improve the coordination performance of a large-scale device cluster.

[0124] In this embodiment, step S300 involves implementing millisecond-level real-time control at the edge control layer to generate pulse width modulation drive signals, and implementing multi-objective global optimization at the cloud optimization layer to dynamically adjust the combination of operating parameters, including the following steps C1-C2:

[0125] C1: The edge control layer executes space vector pulse width modulation drive signal generation, instantaneous power calculation and overmodulation compensation algorithm at a preset period. The cloud optimization layer periodically performs global multi-objective optimization and combines digital twin model to predict operation trend.

[0126] Edge control layer: Deploy a high-speed FPGA controller, executing in 500μs cycles.

[0127] Generate SVPWM drive signal; calculate instantaneous power: P = 1.5 (V d ·i d +v q ·i q ); A modulation compensation algorithm was implemented to suppress harmonics;

[0128] Cloud-based optimization layer: Performs global optimization every 5 minutes, predicts load trends based on a digital twin model, and dynamically adjusts: the number of grid-connected devices; modulation depth δ and switching frequency f. sw Combination; filter parameters (L = 1.2 mH, C = 220 μF).

[0129] Specifically, the edge control layer uses an ARM Cortex-A72 quad-core processor in conjunction with a Xilinx Zynq-7000 FPGA coprocessor to achieve high-performance real-time control. A custom PWM generator IP core is deployed within the FPGA, supporting 32 independent PWM outputs with a 16-bit resolution and a maximum switching frequency of 20kHz. The Space Vector Pulse Width Modulation (SVPWM) algorithm implementation process is as follows: First, a Clarke transform is performed to convert the three-phase voltages into a two-phase stationary coordinate system.

[0130]

[0131] Then determine the sector where the voltage vector is located:

[0132]

[0133] Where, θ=atan2(V β V α ); Calculate the duration of action of the fundamental vector:

[0134]

[0135]

[0136] T0 = ​​T s -T1-T2

[0137] Where T s For the switching period, θ' is the angle of the vector within the sector, and V ref For reference voltage amplitude, V dc The DC bus voltage is used; finally, a PWM drive signal is generated, which adopts a seven-segment symmetrical modulation mode, and the switching sequence is: V0-V1-V2-V7-V2-V1-V0.

[0138] Instantaneous power calculation uses a dq synchronous rotating coordinate system and is achieved through the Park transformation:

[0139] i d =i α ×cosθ+i β ×sinθ

[0140] i q =-i α ×sinθ+i β ×cosθ

[0141] Where θ is the phase angle of the grid voltage output by the phase-locked loop. Active and reactive power are calculated as: P = 1.5(v d ·i d +v q ·i q ), Q = 1.5(v) d ·i d +v q ·i q Power measurement accuracy is achieved through a dual process of hardware and software digital filters: the hardware filter uses a 4th-order Butterworth low-pass filter with a cutoff frequency of 1kHz; the software filter uses a moving average filter with a window length of 20 sampling points.

[0142] The overmodulation compensation algorithm is used to extend the linear modulation range and is activated when the modulation index M > 1. The compensation strategy employs the minimum harmonic injection method: injecting the third harmonic component into the reference voltage.

[0143]

[0144] Expand the modulation index to Overmodulation interval division: 1 < M ≤ 1.155 is the overmodulation region I, and the trajectory correction algorithm is adopted; M > 1.155 is the overmodulation region II, and the six-step wave modulation is adopted. Real-time calculation of the modulation index:

[0145]

[0146] When M approaches the overmodulation boundary, start the compensation algorithm 5 control cycles in advance.

[0147] The digital twin model is constructed based on the physics-informed neural network PINN. The network structure is an 8-layer fully connected neural network with 128 neurons in each layer, and the activation function uses tanh. Physical constraints include basic laws of the power grid such as power balance equations, voltage stability constraints, and frequency stability constraints. The model is trained using the Adam optimizer, with a learning rate of 0.001, a batch size of 256, and 1000 training epochs.

[0148] It should be noted that the PINN network of the digital twin model solves partial differential equations through automatic differentiation technology, avoiding the grid dependence of traditional numerical methods. The network inputs include time t, spatial coordinates (x, y), and system state variables, and the output is the predicted value of the system response. During the training process, the physical constraint loss weight adopts an adaptive adjustment strategy to ensure strict compliance with physical laws.

[0149] C2: Form a closed-loop control through bidirectional data interaction and policy feedback.

[0150] Specifically, the bidirectional data interaction uses Ethernet communication based on the Time-Sensitive Network TSN, supporting deterministic delay and bandwidth guarantee. The network topology adopts a redundant ring structure, with a main loop bandwidth of 1 Gbps and a backup loop bandwidth of 100 Mbps. Time synchronization uses the IEEE1588v2 Precision Time Protocol PTP, with a synchronization accuracy of < 1 μs. The data encapsulation protocol uses the Real-Time Publish-Subscribe Protocol RTPS, supporting multicast communication and Quality of Service QoS management.

[0151] The closed-loop control system adopts a hierarchical structure: device control layer, coordination control layer, and supervision control layer. The device control layer has a cycle of 50 μs, responsible for PWM generation and basic protection; the coordination control layer has a cycle of 500 μs, responsible for power distribution and mode switching; the supervision control layer has a cycle of 5 s, responsible for global optimization and policy distribution. Asynchronous communication is achieved between layers through a message queue, and a priority scheduling algorithm is used to ensure real-time performance: high-priority tasks are preemptively scheduled, and tasks with the same priority are scheduled in a time-slicing manner.

[0152] The performance evaluation indicators for feedback control include: steady-state error ess=lim(t→≦)|r(t)-y(t)|<1%, overshoot σ%<5%, and settling time ts<100ms. The controller parameters are tuned using a particle swarm optimization (PSO) algorithm: the particle dimension equals the number of parameters to be optimized, the search space is set based on engineering experience, and the fitness function uses the integral absolute error (IAE): J=∫|e(t)|dt.

[0153] It should be noted that the design of the closed-loop control system fully considers multi-timescale characteristics and uncertainties. Fast-timescale control ensures system stability, while slow-timescale optimization improves system performance. Robustness is guaranteed by H≦ control theory: ||Tzw||≦<γ, where Tzw is the closed-loop transfer function and γ is the performance index. Adaptive control addresses model uncertainties through online parameter identification: parameter estimation uses the recursive least squares (RLS) algorithm with a forgetting factor λ = 0.995.

[0154] In summary, the distributed edge-sensing design of the multi-source heterogeneous parameter acquisition system enables comprehensive real-time monitoring of device parameters, power grid parameters, and environmental parameters, solving the control blind spot problem caused by traditional single-parameter monitoring. The multi-level filtering architecture, through the combined application of median filtering, Kalman filtering, and wavelet transform, effectively eliminates sensor noise interference and improves data quality reliability. In the dual-layer collaborative architecture, the FPGA high-speed controller in the edge control layer achieves precise control of space vector pulse width modulation, while the multi-objective optimization algorithm in the cloud optimization layer generates globally optimal parameters, forming an organic combination of rapid response and global optimization. The construction of a six-dimensional feature vector based on a hidden Markov model overcomes the limitations of traditional single-dimensional judgment, achieving intelligent prediction of mode switching through a state transition probability matrix, thus shortening response time. The layered design of the three-level fault response mechanism achieves full-coverage protection from early warning to emergency handling. The synergistic effect of dynamic derating algorithms, filter parameter reconstruction, and bypass circuit switching ensures the safe and stable operation of the system under various abnormal conditions. The virtual synchronous machine model maintains grid stability in islanded mode, the dynamic current sharing algorithm achieves power balance in parallel mode, and the multi-dimensional energy efficiency evaluation model optimizes equipment combination in hybrid mode. The adaptive switching of the three operating modes meets the operating requirements under different working conditions.

[0155] Example 3

[0156] Reference Figure 1 - Figure 4 This is the third embodiment of the present invention.

[0157] Parameter settings: The simulation time, number of converters (main converter + backup), mapping relationship between switching frequency and power (assuming that the power of a single unit is proportional to the frequency), and load demand curves (load changes at 0.2s, 1.2s, and 1.7s) are defined.

[0158] Event Trigger: Simulate a primary converter failure at 0.2s and automatically switch to the backup converter. Figure 3 ).

[0159] Core control logic:

[0160] When the power grid is stable: the mode is dynamically switched according to the load demand (single → hybrid → parallel), low frequency is used to save energy when the load is low, and high frequency is used to meet the demand when the load is high.

[0161] When the power grid is unstable: a parallel mode (at least 2 units) is forced to be used to ensure power supply stability through high frequency (third test).

[0162] Dynamically adjust the number of parallel connections: calculate the required number of parallel connections based on the real-time load to ensure that the total output is greater than or equal to the load demand.

[0163] Results visualization: Four sets of curves were plotted, showing the comparison between load and total output, changes in the number of parallel connections, converter status (including fault switching), and switching frequency adjustment, to intuitively verify the effectiveness of the control strategy.

[0164] Simulation verification points:

[0165] Single mode (0-0.2s): Low load (4kW), 1 parallel connection, 500Hz switching frequency (energy saving).

[0166] Hybrid / parallel switching (after 0.2s): When the load increases to 12kW, the number of parallel connections increases to 2 (hybrid mode); when the load further increases to 25kW, the number of parallel connections increases to 3 (parallel mode).

[0167] Fault handling (0.2s): If the main converter 1 fails, the standby converter will be put into operation immediately to ensure the stability of the parallel connection. Load reduction and adjustment (after 1.7s): The load will be reduced to 10kW, and the number of parallel connections will be reduced to 2 to avoid power redundancy.

[0168] Unstable power grid (after 1.5s): Even if the load decreases, the number of parallel connections remains ≥2 to ensure power supply stability.

[0169] Example 4

[0170] The above is a schematic scheme of a collaborative optimization control method for a multimodal power conversion device. It should be noted that the technical solution of the collaborative optimization control system for this multimodal power conversion device belongs to the same concept as the technical solution of the collaborative optimization control method for the multimodal power conversion device described above. Details not described in detail in the technical solution of the collaborative optimization control system for the multimodal power conversion device in this embodiment can be found in the description of the technical solution of the collaborative optimization control method for the multimodal power conversion device described above.

[0171] This embodiment also provides a collaborative optimization control system for a multi-mode power conversion device, including:

[0172] The multi-source data acquisition module collects multi-source heterogeneous operating parameters and constructs a two-layer collaborative architecture to obtain information on device operating status, power grid dynamic characteristics and environmental changes. It establishes a cross-level communication protocol to achieve data synchronization and outputs real-time operating condition feature vectors.

[0173] The module is built upon a dynamic mode switching engine deployed based on real-time operating condition feature vectors. It triggers multi-modal operation strategy reconstruction through a state transition probability model, constructs an adaptive switching mechanism that includes parallel power supply mode, islanded operation mode, and hybrid modulation mode, and establishes a composite fault response system to achieve pre-diagnosis and graded handling of abnormal states.

[0174] The parameter adjustment module performs millisecond-level real-time control at the edge control layer to generate pulse width modulation drive signals, and performs multi-objective global optimization at the cloud optimization layer to dynamically adjust the combination of operating parameters, forming a closed-loop control through bidirectional data interaction and strategy feedback.

[0175] This embodiment also provides an electronic device suitable for the collaborative optimization control of a multi-mode power conversion device, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the collaborative optimization control method for a multi-mode power conversion device proposed in the above embodiment.

[0176] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the collaborative optimization control method for implementing a multi-mode power conversion device proposed in the above embodiments.

[0177] The storage medium proposed in this embodiment and the collaborative optimization control method for realizing a multi-mode power conversion device proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0178] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0179] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A collaborative optimization control method for a multi-mode power conversion device, characterized in that: include, Collect multi-source heterogeneous operating parameters and construct a two-layer collaborative architecture to obtain information on device operating status, power grid dynamic characteristics and environmental changes, establish a cross-level communication protocol to achieve data synchronization, and output real-time operating condition feature vectors; Based on real-time operating condition feature vectors, a dynamic mode switching engine is deployed. Multi-modal operation strategy reconstruction is triggered by state transition probability model. An adaptive switching mechanism including parallel power supply mode, islanded operation mode and hybrid modulation mode is constructed. A composite fault response system is established to realize the pre-diagnosis and graded handling of abnormal states. Millisecond-level real-time control is implemented at the edge control layer to generate pulse width modulation drive signals, and multi-objective global optimization is implemented at the cloud optimization layer to dynamically adjust the combination of operating parameters. Closed-loop control is formed through bidirectional data interaction and strategy feedback.

2. The collaborative optimization control method for a multi-mode power conversion device as described in claim 1, characterized in that: The multi-source heterogeneous operating parameters include device parameters, power grid parameters, and environmental parameters. Device parameters include modulation depth, switching frequency, and junction temperature monitoring. Power grid parameters include voltage imbalance, frequency deviation, and load rate changes. Environmental parameters include wind speed and light intensity changes. The constructed dual-layer collaborative architecture includes an edge control layer and a cloud optimization layer. The edge control layer performs spatial vector pulse width modulation signal generation and instantaneous power calculation, while the cloud optimization layer generates a globally optimal set of weight coefficients through a multi-objective optimization algorithm.

3. The collaborative optimization control method for a multi-mode power conversion device as described in claim 2, characterized in that: The dynamic mode switching engine based on real-time operating condition feature vector deployment includes constructing a multi-dimensional feature vector containing load rate, frequency deviation, harmonic distortion rate, power factor, junction temperature and state of charge based on a hidden Markov model, and triggering mode switching decisions through a state transition probability matrix. Establish a three-level fault response mechanism: early warning level dynamic derating, protection level filter parameter reconfiguration, and emergency level bypass circuit switching.

4. The collaborative optimization control method for a multi-mode power conversion device as described in claim 3, characterized in that: The dynamic mode switching engine triggers a mode switching decision tree based on the entry of a multi-dimensional feature vector into a critical region, thereby controlling the power wave during the switching process. The dynamic impedance matching algorithm is implemented by calculating the equivalent impedance in real time and adjusting the filter parameters, using a repetitive controller to compensate for harmonic components, and adjusting the switching frequency according to the junction temperature distribution.

5. The collaborative optimization control method for a multi-mode power conversion device as described in claim 4, characterized in that: The multi-modal operation strategy reconfiguration includes: in the parallel power supply mode, power balance distribution of multiple devices is achieved through dynamic current sharing algorithm, and redundancy mechanism is adopted to ensure high load conditions. The islanded operation mode maintains grid voltage and frequency stability through a virtual synchronous machine model and autonomously switches operating states when the voltage at the point of common coupling is abnormal. The hybrid modulation mode dynamically selects the optimal combination of equipment through a multi-dimensional energy efficiency assessment model, and adjusts the number of grid-connected devices and modulation depth according to changes in load rate.

6. The collaborative optimization control method for a multi-mode power conversion device as described in claim 5, characterized in that: The established composite fault response system triggers a dynamic derating algorithm by monitoring junction temperature, reconstructs filter parameters by detecting harmonic distortion rate, and triggers bypass circuit switching by monitoring DC bus voltage.

7. The collaborative optimization control method for a multi-mode power conversion device as described in claim 6, characterized in that: It also includes an edge control layer that executes space vector pulse width modulation drive signal generation, instantaneous power calculation and overmodulation compensation algorithms at a preset period, and a cloud optimization layer that periodically performs global multi-objective optimization and combines digital twin models to predict operating trends.

8. A collaborative optimization control system for a multi-mode power conversion device, based on the collaborative optimization control method for the multi-mode power conversion device according to any one of claims 1 to 7, characterized in that: It also includes a multi-source data acquisition module, which collects multi-source heterogeneous operating parameters and builds a two-layer collaborative architecture to obtain information on device operating status, power grid dynamic characteristics and environmental changes, establishes a cross-level communication protocol to achieve data synchronization, and outputs real-time operating condition feature vectors. The module is built upon a dynamic mode switching engine deployed based on real-time operating condition feature vectors. It triggers multi-modal operation strategy reconstruction through a state transition probability model, constructs an adaptive switching mechanism that includes parallel power supply mode, islanded operation mode, and hybrid modulation mode, and establishes a composite fault response system to achieve pre-diagnosis and graded handling of abnormal states. The parameter adjustment module performs millisecond-level real-time control at the edge control layer to generate pulse width modulation drive signals, and performs multi-objective global optimization at the cloud optimization layer to dynamically adjust the combination of operating parameters, forming a closed-loop control through bidirectional data interaction and strategy feedback.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the collaborative optimization control method for the multi-mode power conversion device according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the collaborative optimization control method for the multi-mode power conversion device according to any one of claims 1 to 7.

Images