Pfc system real-time energy efficiency optimization method and system based on dynamic weight particle swarm algorithm

Abstract

The application discloses a kind of PFC system real-time energy efficiency optimization method and system based on dynamic weight particle swarm algorithm, comprising: first, the multidimensional parameter search space of PFC system is constructed, and system operating state parameter is collected in real time. Then, initialize particle swarm population, and iteration optimization is carried out using dynamic weight particle swarm algorithm. Algorithm is updated particle position and speed by dynamically adjusting inertia weight, combining particle own historical optimum and population global optimum information, and finally finds global optimum energy efficiency parameter combination. Finally, according to the combination generation control instruction and issue to each execution unit of PFC system, realize the real-time adjustment of parameter and the closed-loop optimization of system energy efficiency. The application effectively improves the optimization speed and global optimization ability, and guarantees the efficient and stable operation of PFC system.

Description

Technical Field

[0001] This invention relates to the field of PFC system management, and more specifically, to a method and system for real-time energy efficiency optimization of PFC systems based on dynamic weighted particle swarm optimization algorithm. Background Technology

[0002] In the field of power electronics, power factor correction (PFC) systems are crucial for improving power quality and equipment operating efficiency. Traditional PFC system energy efficiency optimization often relies on factory-preset fixed parameters or offline optimization based on typical operating conditions, making it difficult to adapt to the challenges posed by dynamic changes in load and grid conditions during actual operation. Existing online optimization methods, such as optimization algorithms using fixed parameters, often suffer from slow convergence speeds and are prone to getting trapped in local optima in complex multidimensional parameter spaces, resulting in the inability to track and maintain the system's optimal energy efficiency point in real time. Summary of the Invention

[0003] The purpose of this invention is to provide a real-time energy efficiency optimization method and system for PFC systems based on dynamic weighted particle swarm optimization algorithm.

[0004] In a first aspect, embodiments of the present invention provide a real-time energy efficiency optimization method for a PFC system based on a dynamic weighted particle swarm optimization algorithm, comprising: Constructing a search space for energy efficiency optimization parameters of PFC systems; Obtain the set of real-time energy efficiency status parameters of the PFC system under its current operating conditions; The particle swarm population is initialized based on the search space of the PFC system energy efficiency optimization parameters and the set of real-time energy efficiency status parameters to obtain an initial distribution set of the particle swarm. The initial distribution set of the particle swarm contains multiple particle units. Each particle unit has an initial position parameter vector and an initial velocity parameter vector in the search space of the PFC system energy efficiency optimization parameters. The initial position parameter vector represents a set of control parameters to be executed by the PFC system. The dynamic weighted particle swarm optimization algorithm is used to perform multiple rounds of iterative optimization on the initial distribution set of the particle swarm. By updating the particle velocity and position based on historical optimal, global optimal and dynamic inertia weight, the global optimal parameters are output as the optimal energy efficiency parameter combination of the PFC system. A set of real-time energy efficiency optimization control instructions for the PFC system is generated based on the optimal energy efficiency parameter combination of the PFC system, and the set of real-time energy efficiency optimization control instructions for the PFC system is sent to the corresponding adjustable execution unit in the PFC system to trigger the parameter adjustment operation of the execution unit.

[0005] In a second aspect, embodiments of the present invention provide a server system, including a server, for performing the method described in the first aspect.

[0006] Compared to existing technologies, the beneficial effects provided by this invention include: The real-time energy efficiency optimization method and system for PFC systems based on dynamic weighted particle swarm optimization (PSO) provided in this embodiment constructs a multi-dimensional parameter search space for the PFC system and collects system operating status parameters in real time. Next, the particle swarm population is initialized, and iterative optimization is performed using the dynamic weighted PSO algorithm. The algorithm updates particle positions and velocities by dynamically adjusting inertia weights, combining the particle's historical optimal information with the global optimal information of the population, ultimately finding the globally optimal combination of energy efficiency parameters. Finally, control commands are generated based on this combination and sent to each execution unit of the PFC system, achieving real-time parameter adjustment and closed-loop optimization of system energy efficiency. This invention effectively improves optimization speed and global optimization capability, ensuring the efficient and stable operation of the PFC system. Attached Figure Description

[0007] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as limiting the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a flowchart illustrating the steps of a real-time energy efficiency optimization method for a PFC system based on a dynamic weighted particle swarm optimization algorithm, as provided in an embodiment of the present invention. Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0010] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0011] In order to solve the technical problems mentioned in the background art Figure 1 This is a flowchart illustrating the real-time energy efficiency optimization method for a PFC system based on the dynamic weighted particle swarm optimization algorithm provided in this embodiment. The following is a detailed description of this real-time energy efficiency optimization method for a PFC system based on the dynamic weighted particle swarm optimization algorithm.

[0012] Step S201: Construct the search space for energy efficiency optimization parameters of the PFC system; Step S202: Obtain the set of real-time energy efficiency status parameters under the current operating conditions of the PFC system; Step S203: Initialize the particle swarm population according to the PFC system energy efficiency optimization parameter search space and the real-time energy efficiency state parameter set to obtain the initial distribution set of the particle swarm. The initial distribution set of the particle swarm contains multiple particle units. Each particle unit has an initial position parameter vector and an initial velocity parameter vector in the PFC system energy efficiency optimization parameter search space. The initial position parameter vector represents a set of control parameters to be executed by the PFC system. Step S204: The dynamic weighted particle swarm algorithm is used to perform multiple rounds of iterative optimization on the initial distribution set of the particle swarm. By updating the particle velocity and position based on the historical optimal, global optimal and dynamic inertia weight, the global optimal parameters are output as the optimal energy efficiency parameter combination of the PFC system. Step S205: Generate a set of real-time energy efficiency optimization control instructions for the PFC system based on the optimal energy efficiency parameter combination of the PFC system, and send the set of real-time energy efficiency optimization control instructions for the PFC system to the corresponding adjustable execution unit in the PFC system to trigger the parameter adjustment operation of the execution unit.

[0013] In this embodiment of the invention, exemplarily, the server performs a build operation. Specifically, the server first collects the set of device specification parameters for all dynamically adjustable actuators in the PFC system through its data interface. For example, in a three-phase PFC rectifier system, these adjustable actuators may include dead-time regulators for the switching transistor drive signals of each phase arm, Kp and Ki parameter regulators for the current loop proportional-integral (PI) controller, and modulation ratio limiters for space vector pulse width modulation (SVPWM). The server reads the specification parameters of each regulator from the device configuration database, including its adjustable lower limit value (e.g., a minimum dead time of 100 nanoseconds) and upper limit value (e.g., a maximum dead time of 500 nanoseconds). Based on these upper and lower limit values, the server defines a dimensional boundary interval for each adjustable parameter, for example, the dead time dimensional interval is [100ns, 500ns], and the Kp parameter dimensional interval is [0.1, 5.0].

[0014] Subsequently, the server extracts a set of linkage constraint relationships between these adjustable parameters from the system constraint configuration file. For example, the constraint might stipulate that the current loop PI parameters of phase A and phase B must be adjusted proportionally and synchronously, and their Kp values ​​must be maintained at a ratio of 1:1.2; or it might stipulate that the modulation ratio adjustment must be performed after the dead time adjustment is completed. Based on this set of parameter linkage constraint relationships, the server performs correlation correction processing on the defined boundary intervals of each dimension. For example, once the Kp value of phase A is determined, the effective interval of the Kp value of phase B will dynamically shrink according to the proportional constraint, thereby generating a multi-dimensional coupled parameter search space that satisfies all parameter linkage constraint conditions. In this space, the boundaries of different dimensions are no longer independent, but form coupled and correlated edges through constraint relationships.

[0015] Next, the server performs spatial discretization and grid partitioning on the aforementioned multidimensional coupled parameter search space. The server reads preset grid partitioning density parameters, for example, dividing each dimension into 10 equal parts. The server divides the dead time dimension interval [100ns, 500ns] into 10 sub-interval units, each with a length of 40ns and center values ​​of 120ns, 160ns, ..., 480ns respectively. Similarly, other parameter dimensions are partitioned in a similar manner. Through this operation, the server generates a discretized parameter search grid space with clearly defined grid node coordinates. The server extracts the node coordinate vector of each grid node; for example, a node coordinate vector might be (dead time = 200ns, Kp_A = 1.5, Ki_A = 0.05, Kp_B = 1.8, ...). Based on these node coordinate vectors, the server constructs an index mapping table for the PFC system energy efficiency optimization parameter search space, associating each unique coordinate vector with an index ID (e.g., ID_001). The server combines all index IDs to generate a discretized search space topology for subsequent particle swarm optimization. The server also marks the adjacency relationships between each grid node in this topology. For example, in the dead-time dimension, the node at coordinate 200ns is adjacent to the nodes at coordinates 160ns and 240ns. Finally, the server generates a sparse grid node connectivity matrix based on these adjacency relationships, which is used to understand the local structure of the search space in subsequent algorithms.

[0016] The server periodically acquires real-time data from various sensor networks connected to the PFC system. It collects instantaneous waveform sequences of the input-side voltage and current from the power quality monitoring sensor array installed at the PFC system input. The server performs waveform feature extraction processing on the input-side voltage waveform sequences, including Fast Fourier Transform (FFT) and RMS value calculation, to obtain the input-side voltage RMS parameters, total harmonic distortion (THD) parameters, and input-side voltage frequency fluctuation parameters. These collectively constitute part of the power quality monitoring parameters. Simultaneously, the server performs the same processing on the input-side current waveform sequences to obtain the input-side current RMS value, current THD, and the phase difference (power factor angle) between the current and voltage.

[0017] The server collects instantaneous voltage and current waveform sequences on the output side after compensation from the power quality compensation sensor array at the PFC system output. The server evaluates the compensation effect of these waveform sequences, calculating the deviation between the compensated waveform and the ideal sine wave to obtain the output-side voltage harmonic compensation residual parameter and the output-side voltage amplitude stability parameter. Simultaneously, by analyzing the phase relationship between the output-side current waveform and the voltage waveform, the output-side power factor compensation compliance parameter is calculated.

[0018] In addition, the server acquires instantaneous junction temperature sequences of semiconductor devices and instantaneous temperature sequences of magnetic components (inductors and transformers) from temperature monitoring sensor arrays within the PFC system's power conversion units (such as IGBT modules, inductors, and transformers). The server also acquires instantaneous DC bus voltage sequences and instantaneous voltage sequences across the switching transistors from voltage monitoring points (acquired via ADCs) in the power conversion units. Furthermore, the server acquires instantaneous on-state current sequences of the switching transistors and instantaneous inductor current sequences from current monitoring points (such as Hall effect sensors).

[0019] Based on these high-frequency raw data streams, the server performs real-time calculations. Using the instantaneous sequences of semiconductor device junction temperatures, voltages across switching transistors, and on-current of switching transistors, combined with the loss models from the device datasheets, the server calculates the instantaneous switching losses of semiconductor devices (such as IGBTs and diodes) in the power conversion unit. Simultaneously, based on the instantaneous sequences of magnetic component temperatures, inductor currents, and DC bus voltages, combined with the loss curves of magnetic materials (such as the Steinmetz equation) and winding resistance, the server calculates the instantaneous iron and copper losses of the magnetic components. All these loss values ​​together constitute the loss status parameters. Finally, the server aggregates the power quality monitoring parameters, power quality compensation parameters, and loss status parameters to form a real-time energy efficiency status parameter set describing the current energy efficiency status of the PFC system.

[0020] After receiving the constructed search space and real-time state parameters, the server begins initializing the particle swarm optimization algorithm. First, the server parses the set of dimensional boundary intervals in the PFC system energy efficiency optimization parameter search space, obtaining the lower and upper limits of adjustment for each control parameter dimension (a total of D dimensions). The server then reads the preset particle population size parameter N (e.g., N=50).

[0021] The server generates N uniformly distributed random number sequences of length D. For the i-th particle unit, the server linearly maps the d-th random number in its corresponding random number sequence to the interval formed by the lower limit Ld and the upper limit Ud of the d-th control parameter dimension, calculated as: Initial position component = Ld + rand()*(Ud - Ld). By traversing all dimensions, the server generates the initial position parameter vector for this particle unit, which represents a set of random combinations of PFC system control parameters.

[0022] Next, the server generates a sequence of random velocity components with the same dimension as the position vector for each particle unit based on the preset initial velocity range parameters (for example, the velocity range is set to 20% of the corresponding position dimension range). The value of each velocity component is randomly generated within the preset velocity range and serves as the initial velocity parameter vector for that particle unit.

[0023] Subsequently, the server inputs the initial position parameter vector of each particle unit, along with the current real-time energy efficiency state parameter set, into a preset PFC system energy efficiency evaluation function for calculation. This evaluation function may be a weighted model that integrates indicators such as input power factor, output THD, and total losses. The server obtains the initial energy efficiency evaluation value (e.g., an efficiency percentage or a comprehensive score) for each particle unit and uses this value as the initial fitness value for that particle unit.

[0024] During initialization, the server sets the initial position parameter vector of each particle unit to its historical best position parameter vector and its initial fitness value to its historical best fitness value. Then, the server compares the initial fitness values ​​of all N particle units, selecting the particle unit with the best fitness value (e.g., highest efficiency) and marking it as the initial globally optimal particle unit. The server then uses the initial position parameter vector of this optimal particle as the globally optimal position parameter vector for the entire particle swarm and its initial fitness value as the globally optimal fitness value.

[0025] Finally, the server creates a data structure for each particle unit, containing its identifier, current round position / velocity vector, and individual historical best position / fitness value. The server aggregates the state structures of all particle units to form a complete initial distribution set of particles containing all state information of the population, and stores it in memory for subsequent iterations.

[0026] The server invokes the dynamic weighted particle swarm optimization (PSO) module to iteratively optimize the initial distribution set of the particle swarm. The server first sets the current iteration round counter t and initializes it to 1. The server then reads the preset maximum iteration round threshold parameter T_max (e.g., T_max = 100).

[0027] At the start of each iteration, the server extracts the current-cycle position parameter vector and current-cycle velocity parameter vector of all particle units from the particle swarm distribution set in memory. The server then calculates the dynamic weights based on the current iteration counter t and a preset dynamic weight formula (e.g., w = w_max - (w_max - w_min) * (t / T_max)). 2 The algorithm calculates the dynamic inertia weight coefficient w corresponding to the current iteration, where w_max represents the preset maximum inertia weight (e.g., 0.9) and w_min represents the preset minimum inertia weight (e.g., 0.4). This coefficient decreases non-linearly from a large value to a small value as the iteration progresses. This non-linear decrease in coefficient as the iteration progresses allows the algorithm to focus on global exploration in the early stages and local development in the later stages.

[0028] For each particle unit i in the population, the server calculates v according to the standard particle swarm velocity update formula: i (t+1)=w·v i (t)+c1·r1·(p i (t)-x i (t)+c2·r2·(p g (t)-x i (t)). Wherein, v i (t+1) is the uncorrected velocity vector of the i-th particle unit in the (t+1)-th iteration, w is the dynamic inertia weight coefficient corresponding to the current iteration, and v i (t) represents the current velocity vector of the i-th particle unit in the t-th round, c1 and c2 are preset learning factors, r1 and r2 are mutually independent uniformly distributed random numbers in the interval [0,1], and p i (t) represents the historical best position vector of an individual from the i-th particle unit up to the t-th round, x i (t) is the current position vector of the i-th particle unit in the t-th round, p g(t) represents the globally optimal position vector of the entire particle swarm up to round t. The server generates the updated velocity parameter vector for this particle unit.

[0029] To prevent excessive speed from causing search instability, the server updates the velocity vector v for each particle unit. i Each velocity component of (t+1) undergoes velocity boundary constraint processing. For example, if a velocity component exceeds the preset upper velocity limit V_max, the server restricts it to V_max; if it is below the lower velocity limit V_min, it is restricted to V_min, thus obtaining the corrected velocity parameter vector of the particle unit.

[0030] Next, the server updates the formula based on the location: x i (t+1)=x i (t)+v' i (t+1), generate the updated position parameter vector for this particle unit, where x i (t+1) is the uncorrected position vector of the i-th particle unit in the (t+1)-th round, x i (t) is the current position vector of the i-th particle unit in the t-th round, v' i (t+1) represents the corrected velocity vector of the i-th particle unit after velocity boundary constraints in the (t+1)-th round. Similarly, to prevent the position from exceeding the feasible parameter adjustment range, the server updates the position vector x... i Each position component of (t+1) undergoes position boundary constraint processing. For example, if the calculated value of the dead time component is 550ns, which exceeds the adjustment limit of 500ns for this dimension, the server limits it to 500ns, resulting in the corrected position parameter vector.

[0031] Then, the server inputs the corrected position parameter vector (i.e., a new set of control parameters) of each particle unit into the PFC system energy efficiency evaluation function for calculation, and obtains the energy efficiency evaluation value of the particle unit in the current round, which is used as the fitness value of the current round.

[0032] The server compares the current fitness value of a particle with its stored historical best fitness value. If the current fitness value is better (e.g., more efficient), the server updates the particle's corrected position parameter vector to the particle's historical best position parameter vector and updates the current fitness value to the historical best fitness value.

[0033] After completing the individual updates of all particle units, the server compares the current fitness value of all particle units with the currently stored global optimal fitness value of the particle swarm. The server checks if there is at least one particle unit whose current fitness value is better than the global optimal fitness value. If so, the server selects the particle with the best fitness value from these better particles and marks it as the global optimal particle unit for this round. The server updates the corrected position parameter vector of this particle to the global optimal position parameter vector of the particle swarm and updates its fitness value to the global optimal fitness value. Then, the server increments the value of the current iteration round counter t by 1. The server determines whether the updated t value has reached the preset maximum iteration round threshold T_max. If not, the server returns to the "calculate dynamic inertia weight coefficient" step and begins the next round of iteration optimization, continuing to optimize particle positions based on the updated individual optimal and global optimal information. If T_max has been reached, the server terminates the iteration optimization process.

[0034] The server extracts the final globally optimal position parameter vector from the particle swarm state at termination. This vector represents the theoretically optimal combination of control parameters for the PFC system's overall energy efficiency under the current operating conditions, after multiple rounds of iterative optimization. The server outputs this vector as the optimal energy efficiency parameter combination for the PFC system.

[0035] After obtaining the optimal energy efficiency parameter combination, the server begins to generate and issue control commands. The server first parses the parameter combination, where each parameter component corresponds to a target adjustable execution unit in the PFC system (such as "dead time regulator" or "A-phase Kp regulator").

[0036] The server queries and retrieves the real-time operating parameter values ​​of each adjustable execution unit in the PFC system via the real-time data bus. For each target adjustable execution unit, the server calculates the difference between its optimal parameter component value and the current real-time operating parameter value to obtain the parameter adjustment deviation.

[0037] Based on the sign and magnitude of the deviation, the server determines the adjustment direction identifier (e.g., "increase" or "decrease") and adjustment percentage (e.g., "increase by 5%) for the execution unit. Subsequently, the server encapsulates the adjustment direction identifier and adjustment percentage into a specific unit control command data packet that can be recognized by the lower-level controller, according to a predefined device control protocol format (e.g., Modbus RTU message, CAN bus frame) that matches each execution unit.

[0038] To improve the orderliness and reliability of control, the server adds a timestamp (current system time) and a priority flag to each generated unit control instruction data packet. The priority is pre-set based on the importance of the execution unit in the system and the urgency of the adjustment; for example, DC bus voltage regulation instructions, which are related to system stability, have the highest priority.

[0039] The server sorts all tagged unit control instruction packets according to their instruction priority from high to low, generating an instruction execution queue with a clear execution order. The server then extracts unit control instruction packets sequentially, starting from the head of this queue.

[0040] The server sends the extracted data packets to the digital control interface (such as the PWM register of a DSP or the configuration interface of an FPGA) of the corresponding target adjustable execution unit via the control bus (such as Ethernet or CAN bus) within the PFC system. After sending each instruction data packet, the server starts a timer to wait for and receive an "instruction execution confirmation" signal from the target execution unit.

[0041] After receiving confirmation signals from all target execution units, or after a timeout, the server aggregates this status information and generates an execution result feedback report for the PFC system's real-time energy efficiency optimization control commands. The report details the issuance time, target object, expected value, and confirmation status of each command. Finally, the server stores this execution result feedback report in the PFC system's historical command execution record database for subsequent performance analysis and audit trails. Through this series of operations, the server completes a closed-loop real-time energy efficiency optimization control process from perception and optimization to execution.

[0042] In this embodiment of the invention, the dynamic weighted particle swarm algorithm is used to perform multiple rounds of iterative optimization on the initial distribution set of the particle swarm. By updating the particle velocity and position based on historical optimal, global optimal and dynamic inertia weights, the globally optimal parameters are output as the optimal energy efficiency parameter combination of the PFC system. This can be implemented through the following example.

[0043] The dynamic weighted particle swarm optimization algorithm is invoked to perform multiple rounds of iterative optimization on the initial distribution set of the particle swarm. In each iteration, the velocity parameter vector and position parameter vector of the particle unit are updated according to the historical optimal position parameter vector of the particle unit in the search space of the energy efficiency optimization parameters of the PFC system, the global optimal position parameter vector of the particle swarm population, and the dynamic inertia weight coefficient corresponding to the current iteration round. When the preset iteration termination condition is reached, the global optimal position parameter vector of the particle swarm population is output as the optimal energy efficiency parameter combination of the PFC system.

[0044] In this embodiment of the invention, for example, the server invokes the dynamic weighted particle swarm optimization algorithm module to begin iterative optimization of the initialized particle swarm distribution set. The server first sets the current iteration counter t=1 and reads the preset maximum iteration count T_max=100 as the termination condition. In each iteration, the server extracts the current position vector and velocity vector of all particle units from memory. Based on the current iteration count t, the server uses the formula w=0.9-(0.9-0.4)*(t / 100) 2 The dynamic inertia weight coefficient w for the current round is calculated, and this coefficient decreases nonlinearly from 0.9 to 0.4 as the iteration progresses.

[0045] For each particle in the population, the server performs a standard velocity update formula calculation based on its individual historical best position vector, the current global best position vector of the entire population, and the calculated dynamic inertia weight w, generating a new velocity vector. The server then performs boundary checks on each component of the new velocity vector; if a component exceeds the preset range [-V_max, V_max], it is restricted to the boundary value. Next, the server calculates the particle's new position vector based on the updated velocity vector. The server also performs boundary processing on each component of the new position vector to ensure it does not exceed the allowable adjustment range of the corresponding control parameters (such as dead time and PI parameters).

[0046] The server inputs the updated position vector of each particle (representing a new combination of control parameters) into the PFC system's energy efficiency evaluation function for calculation, obtaining the particle's energy efficiency evaluation value under the current parameters as its fitness for this round. The server compares each particle's fitness for this round with its own historical best fitness; if it is better, it updates its individual historical best position and fitness. Subsequently, the server traverses the entire population, comparing the fitness of all particles for this round with the current global best fitness; if it finds a better particle, it updates the global best position vector and fitness value.

[0047] After completing all the above operations, the server increments the iteration counter t by 1. The server then checks whether t has reached T_max. If not, the server returns to the step of calculating the dynamic inertia weight and begins the next iteration. If the maximum number of iterations has been reached, the server terminates the iteration process and extracts the finally determined global optimal position parameter vector from memory, outputting it as the optimal energy efficiency parameter combination for the PFC system.

[0048] In this embodiment of the invention, the PFC system energy efficiency optimization parameter search space is composed of multiple control parameter dimensions that can be dynamically adjusted during the operation of the PFC system. Each control parameter dimension corresponds to the adjustment range of the adjustable execution unit in the PFC system. The construction of the PFC system energy efficiency optimization parameter search space can be implemented through the following example.

[0049] The server performs the operation of constructing the energy efficiency optimization parameter search space for the PFC system. The server first collects the set of equipment specification parameters for all adjustable actuators from the equipment configuration database. For example, it reads that the lower limit of adjustment for the "dead time regulator" is 100 nanoseconds and the upper limit is 500 nanoseconds, and the lower limit of adjustment for the "A-phase current loop Kp regulator" is 0.1 and the upper limit is 5.0.

[0050] Based on these upper and lower limits, the server defines dimensional boundary intervals for each adjustable parameter. For example, the dead time dimension interval is [100ns, 500ns], and the Kp_A dimension interval is [0.1, 5.0]. Next, the server extracts a set of parameter linkage constraints from the system constraint configuration file. For example, one constraint stipulates that "the adjustment ratio of the Kp value of phase A to the Kp value of phase B must be maintained at 1:1.2", and another constraint stipulates that "the modulation ratio parameter adjustment must be performed after the dead time adjustment is completed".

[0051] The server performs correlation correction processing on the defined boundary intervals of each dimension based on these linkage constraints. For example, when the Kp value of phase A is determined to be 2.0, according to the proportional constraint, the effective interval of the Kp value of phase B is dynamically corrected to [2.4, 6.0], and the intersection with the original upper limit is taken, thereby generating a multi-dimensional parameter search space in which each dimension is coupled with each other.

[0052] Subsequently, the server discretizes the coupled space into a grid. Based on preset grid density parameters (e.g., dividing each dimension into 10 equal parts), the server divides the dead-time dimension [100ns, 500ns] into 10 sub-intervals, each 40ns in length, with center values ​​of 120ns, 160ns, ..., 480ns respectively. After performing similar operations on all dimensions, the server generates a discretized parameter search grid space.

[0053] The server extracts the coordinate vector of each grid node. For example, a node's coordinate vector might be (dead time = 200ns, Kp_A = 1.5, Kp_B = 1.8). Based on these coordinate vectors, the server constructs an index mapping table, associating each unique coordinate vector with a spatial index identifier, such as ID_001. The server combines all index identifiers to generate a discretized search space topology.

[0054] The server marks the adjacency relationships between grid nodes in this topology. For example, the node corresponding to index ID_001 (dead time = 200ns, Kp_A = 1.5) is adjacent to indices ID_002 (dead time = 240ns, Kp_A = 1.5) and ID_003 (dead time = 160ns, Kp_A = 1.5) in the dead time dimension. Finally, the server generates a sparse grid node connectivity matrix based on these adjacency relationships. The rows and columns of the matrix are index identifiers; if two nodes are adjacent, the corresponding matrix element has a value of 1, otherwise it is 0.

[0055] In this embodiment of the invention, the acquisition of the real-time energy efficiency status parameter set under the current operating conditions of the PFC system, the real-time energy efficiency status parameter set including the power quality monitoring parameters at the input end of the PFC system, the power quality compensation parameters at the output end of the PFC system, and the loss status parameters of the power conversion unit inside the PFC system, can be implemented through the following example.

[0056] First, the server acquires raw waveform data from the power quality monitoring sensor array at the input of the PFC system. The server, through its integrated multi-channel synchronous data acquisition card, sends synchronous sampling commands to the voltage transformers (PTs) and current transformers (CTs) installed at the grid connection points. The server simultaneously captures a complete power frequency cycle (20 milliseconds) of the instantaneous input-side voltage waveform sequence (e.g., an array containing 200 data points) and the instantaneous input-side current waveform sequence at a sampling rate of 10,000 times per second (10 kHz). This raw waveform data is then transmitted in real-time to the server's memory buffer via a high-speed bus.

[0057] Next, the server performs waveform feature extraction processing on the input waveform data. The server calls its built-in signal processing algorithm library to perform a Fast Fourier Transform (FFT) on the voltage waveform sequence in the buffer. The server obtains the effective value parameter of the input voltage, for example, 230.5 volts, by calculating the root mean square of the amplitude of the fundamental (50Hz) component. The server calculates the total harmonic distortion (THD) parameter of the input voltage by comparing the square root of the sum of the squares of the amplitudes of each harmonic (e.g., 3rd, 5th, and 7th) with the fundamental amplitude, for example, 3.5%. Simultaneously, the server calculates the frequency fluctuation parameter of the input voltage, for example, 49.98 Hz, by detecting the zero-crossing time interval of the voltage waveform. The server performs the same FFT analysis on the current waveform sequence to obtain the effective value parameter of the input current (e.g., 150.2 amperes) and the current THD parameter (e.g., 25.1%). In addition, the server performs cross-correlation analysis or fundamental phase extraction on voltage and current waveform sequences at the same timestamp, and accurately calculates the phase difference parameter between the two, such as a power factor angle of 38.7 degrees, which is converted to a power factor parameter of 0.78.

[0058] Then, the server collects data from the compensation effect monitoring sensor array at the PFC system output. Similarly, the server uses a data acquisition card to simultaneously acquire the instantaneous voltage and current waveform sequences after compensation from the monitoring PT and CT connected to the load side. The server performs compensation effect evaluation processing on the output voltage waveform. The server again uses FFT analysis to calculate the residual content of each harmonic in the compensated voltage waveform and compares its sum with the fundamental amplitude to obtain the output voltage harmonic compensation residual parameter, for example, 1.2%. The server also calculates the percentage deviation of the maximum and minimum values ​​of the voltage waveform within one cycle relative to the rated amplitude to obtain the output voltage amplitude stability parameter, for example, ±1%. The server analyzes the output current waveform, calculates its harmonic residual, and obtains the output current harmonic compensation residual parameter. Simultaneously, the server precisely calculates the phase difference between the output current and the fundamental voltage to confirm whether it is close to zero degrees, thus obtaining the output power factor compensation compliance parameter, for example, 0.998.

[0059] Simultaneously, the server collects detailed operating status data from the power conversion unit within the PFC system. The server polls the temperature sensor directly mounted inside the IGBT chip housing via the I2C digital temperature sensor bus to obtain the instantaneous junction temperature sequence of the semiconductor devices in the power conversion unit (e.g., 10 samples over the past second, averaging 85°C). The server also reads data from the thermistor embedded in the PFC inductor core via another SPI bus to obtain the instantaneous temperature sequence of the magnetic components (e.g., averaging 65°C). The server reads data from a high-precision analog-to-digital converter (ADC) module specifically designed to monitor the power circuit: from the voltage divider sampling circuit across the DC bus capacitor to obtain the instantaneous DC bus voltage sequence inside the power conversion unit (e.g., 800V DC); and from the differential probe connected across the IGBT collector and emitter to obtain the instantaneous voltage sequence across the switching transistor. The server also reads data from the current sensing circuit (e.g., using a Hall effect current sensor or a sampling resistor with an isolation amplifier): acquiring the instantaneous conduction current sequence flowing through the IGBT switching transistor and the instantaneous inductor current sequence flowing through the PFC inductor.

[0060] Finally, the server performs real-time loss calculations based on the collected sequence data. The server calculates the energy loss per switching action (e.g., 2 millijoules) by interpolating the instantaneous junction temperature sequence (85°C), the instantaneous voltage sequence (e.g., 600V), and the instantaneous current sequence (e.g., 100A) collected at the moment the switch is turned off, combined with the pre-stored relationship curves between the switching energy (Eon, Eoff) and junction temperature and current of this IGBT model in the database. The server then multiplies this by the real-time switching frequency (e.g., 10kHz) to obtain the average instantaneous switching loss of the semiconductor device (e.g., 20 watts). The server adds the switching loss to the conduction loss calculated based on the on-resistance and current, using this sum as the instantaneous semiconductor device loss value. Finally, the server calculates the instantaneous iron loss of the magnetic component (e.g., 5 watts) by calling pre-stored magnetic material loss coefficients (e.g., Steinmetz equation parameters) based on the instantaneous temperature sequence (65°C), the effective value of the inductor current sequence (e.g., 10A), and the switching frequency harmonic components determined by the instantaneous DC bus voltage sequence. The server calculates the instantaneous copper loss of the magnetic component, for example, 3 watts, based on the DC resistance of the inductor winding (considering temperature rise) and the effective value of the inductor current. The server then aggregates all calculated loss values ​​to form a complete set of loss state parameters. At this point, the server has completed the comprehensive acquisition of power quality monitoring parameters, power quality compensation parameters, and loss state parameters, integrating them to generate a real-time energy efficiency state parameter set for subsequent optimization algorithm input.

[0061] In this embodiment of the invention, the step of initializing the particle swarm population based on the search space of the PFC system energy efficiency optimization parameters and the set of real-time energy efficiency state parameters to obtain the initial distribution set of the particle swarm can be implemented through the following example.

[0062] The server performs the initialization operation of the particle swarm optimization population. First, the server parses the constructed PFC system energy efficiency optimization parameter search space, retrieving the lower and upper limits of adjustment for each control parameter dimension from the data structure in memory. For example, the server reads the boundary of the first dimension (dead time) as [100ns, 500ns], the boundary of the second dimension (A-phase Kp parameter) as [0.1, 5.0], and so on, resolving the boundaries of a total of D=5 dimensions.

[0063] The server reads the preset particle population size parameter N=50. Then, the server calls its random number generator to generate N uniformly distributed random number sequences of length D. Each sequence is an array containing 5 random numbers in the interval [0,1]. For example, the random number sequence generated for particle #1 is [0.32,0.85,0.11,0.63,0.42].

[0064] For particle #1, the server linearly maps each value in its random number sequence to the corresponding adjustment range. Taking the first dimension (dead time) as an example, the server calculates: initial position component = 100ns + 0.32 * (500ns - 100ns) = 228ns. The server performs the same mapping calculation on the remaining four random numbers in the sequence, ultimately generating the initial position parameter vector for particle #1, for example (dead time = 228ns, Kp_A = 4.26, Ki_A = 0.042, Kp_B = 5.12, modulation ratio = 0.79).

[0065] Next, the server generates a sequence of random velocity components for each particle based on a preset initial velocity range parameter (e.g., set to 20% of the position range width in each dimension). For the dead time dimension of particle #1, the position range width is 400ns, and 20% of that is 80ns, so the velocity range is [-80ns, 80ns]. The server randomly generates a velocity value within this range, for example, 15ns. The server performs a similar operation on all dimensions to generate the initial velocity parameter vector for particle #1, for example, (15, -0.21, 0.008, 0.33, -0.05).

[0066] Then, the server inputs the initial position parameter vector of particle #1 (i.e., a specific combination of control parameters) along with the currently acquired set of real-time energy efficiency state parameters into the PFC system's energy efficiency evaluation function. This function is a comprehensive model, which the server uses to calculate the system's estimated comprehensive efficiency under these parameters, for example, obtaining an initial energy efficiency evaluation value of 92.5%. The server records this value as the initial fitness value of particle #1.

[0067] The server sets the initial position parameter vector (228ns, 4.26, ...) of particle #1 to its historical best position parameter vector, and sets the initial fitness value of 92.5% to its historical best fitness value. The server repeats the above process of "position generation - velocity generation - fitness calculation - individual best record" for the remaining 49 particles in the population.

[0068] After all individual particle information has been initialized, the server iterates through and compares the initial fitness values ​​of these 50 particles. The server finds that particle #37 has the highest initial fitness value, at 94.1%. The server marks particle #37 as the initial globally optimal particle unit. The server copies the initial position parameter vector of particle #37 as the globally optimal position parameter vector for the entire particle swarm, and sets its fitness value of 94.1% as the globally optimal fitness value.

[0069] Finally, the server constructs a structured data object (state structure) for each particle. Taking particle #1 as an example, its structure includes: particle unit identifier (ID_001), current round position parameter vector (228ns, 4.26, ...), current round velocity parameter vector (15, -0.21, ...), individual historical best position parameter vector (228ns, 4.26, ...), and individual historical best fitness value (92.5%). The server adds the 50 generated particle unit state structures to a list or array, generating a complete initial distribution set of the particle swarm, and loads it into a specific region of memory, preparing for subsequent iterative optimization algorithms.

[0070] In this embodiment of the invention, the dynamic weighted particle swarm algorithm is invoked to perform multiple rounds of iterative optimization on the initial distribution set of the particle swarm. In each iteration, the velocity parameter vector and position parameter vector of the particle unit are updated according to the historical optimal position parameter vector of the particle unit in the search space of the energy efficiency optimization parameters of the PFC system, the global optimal position parameter vector of the particle swarm population, and the dynamic inertia weight coefficient corresponding to the current iteration round. When the preset iteration termination condition is reached, the global optimal position parameter vector of the particle swarm population is output as the optimal energy efficiency parameter combination of the PFC system. This can be implemented through the following example.

[0071] The server performs multi-round iterative optimization using the dynamic weighted particle swarm optimization algorithm. First, the server sets a counter `t` for the current iteration round in memory and initializes it to 1. The server reads the preset maximum iteration round threshold parameter `T_max=100` from the configuration file, which serves as the primary condition for algorithm termination.

[0072] The first iteration begins (t=1). The server extracts the current-cycle position parameter vector and current-cycle velocity parameter vector for all 50 particle units from the initial particle swarm distribution set already loaded into memory. For example, the extracted current position vector for particle #1 is (228ns, 4.26, 0.042, 5.12, 0.79), and the current velocity vector is (15, -0.21, 0.008, 0.33, -0.05).

[0073] The server calculates the dynamic weights according to the preset formula w=w_max-(w_max-w_min)*(t / T_max) based on the current iteration counter t=1. 2 Perform the calculation, where w_max=0.9, w_min=0.4. Substituting these values, we get w=0.9-(0.9-0.4)*(1 / 100). 2 =0.89995, which is the dynamic inertia weight coefficient for the first iteration.

[0074] Subsequently, the server updates the velocity and position of each particle in the swarm. Taking particle #1 as an example, the server obtains its historical best position vector (initialized as (228ns, 4.26, 0.042, 5.12, 0.79)) and the current global best position vector of the swarm (initialized as the position of particle #37, (190ns, 3.80, 0.038, 4.56, 0.82)). The server updates the velocity according to the standard velocity formula: new velocity vi = w * current velocity vi + c1 * r1 * (individual best pi - current position xi) + c2 * r2 * (global best pg - current position xi), where c1 = 2.0, c2 = 2.0, and r1 and r2 are random numbers within [0,1], for example, 0.5 and 0.3 respectively. The server calculates each component of the velocity vector of particle #1 independently. Taking the dead zone time dimension as an example: New velocity = 0.89995*15 + 2.0*0.5*(228-228) + 2.0*0.3*(190-228) ≈ 13.5 + 0 + (-22.8) = -9.3. The server completes the calculation for all dimensions and obtains the updated velocity parameter vector for particle #1, for example (-9.3, 0.05, 0.015, 0.25, -0.02).

[0075] Next, velocity boundary constraints are applied. The preset velocity upper limit V_max is 20% of the width of the position range in each dimension, and the dead time dimension corresponds to 80ns. The server checks the new velocity vector of particle #1 and finds that all components are within the range [-V_max, V_max], so the corrected velocity parameter vector remains unchanged.

[0076] Then, the position is updated. The server calculates the new position xi according to the formula: new position xi = current position xi + corrected velocity vi. Taking the dead time dimension as an example: new position = 228ns + (-9.3ns) = 218.7ns. After calculating all dimensions, the server obtains the updated position parameter vector of particle #1, for example (218.7ns, 4.31, 0.057, 5.37, 0.77).

[0077] Then, position boundary constraints are applied. The server compares each component of the new position vector with the upper and lower limits of the corresponding control parameter dimension. For example, the dead time component of 218.7 ns remains unchanged within the interval [100 ns, 500 ns]; however, the Kp_B component of 5.37 exceeds its upper limit of 5.0, so the server limits it to 5.0. After processing, the corrected position parameter vector for particle #1 is (218.7 ns, 4.31, 0.057, 5.0, 0.77).

[0078] After completing the position update, the server performs fitness evaluation and update. The server inputs the corrected position parameter vector of particle #1 into the PFC system energy efficiency evaluation function, and calculates its current round energy efficiency evaluation value of 92.8% by combining it with real-time state parameters. This value is then used as the current round fitness value. The server compares this value (92.8%) with the particle's stored historical best fitness value (92.5%). Since 92.8% is better, the server updates particle #1's historical best position parameter vector to (218.7ns, 4.31, 0.057, 5.0, 0.77) and updates its historical best fitness value to 92.8%.

[0079] After all particles have completed their individual updates, the server performs a global optimum update. The server iterates through the current fitness values ​​of all 50 particle units and compares them with the currently stored global optimum fitness value (initialized to 94.1%). The server finds that particle #23's current fitness value is 94.5%, which is better than the current global optimum. The server selects the particle with the highest fitness value among all the better particles, i.e., particle #23, and designates it as the global optimum particle unit for this round. The server updates the global optimum position parameter vector of the particle swarm population to the corrected position vector of particle #23, and updates the global optimum fitness value to 94.5%.

[0080] After completing the first iteration, the server updates the iteration counter. The server increments the value of the counter t for the current iteration round by 1, making it 2. The server then checks whether t=2 reaches T_max=100. Since it does not, the server returns to the "calculate dynamic inertia weight coefficient" step and begins the second iteration. In the second iteration, the server uses the updated individual optimal position and global optimal position (the position of particle #23), along with the new dynamic weight calculated based on t=2 (e.g., 0.8998), to perform velocity updates, position updates, boundary handling, fitness calculations, and optimal value updates for all particles again.

[0081] The server repeats the complete iterative loop described above. As the iteration progresses, the dynamic inertia weight coefficient w decreases non-linearly, and the global optimal position and fitness value may be continuously updated by better particles during the iteration. When the server completes the 100th iteration (i.e., t is updated to 101 and it is determined that T_max has been reached), the server terminates the iterative optimization process.

[0082] The server extracts the final globally optimal position parameter vector from the particle swarm state at termination. For example, this vector is (205ns, 3.95, 0.041, 4.74, 0.81), with a corresponding globally optimal fitness value of 95.7%. The server outputs this final globally optimal position parameter vector as the optimal energy efficiency parameter combination of the PFC system and passes it to the subsequent control command generation module.

[0083] In this embodiment of the invention, the step of performing a velocity update operation on each particle unit based on the particle unit's current round velocity parameter vector, the particle unit's individual historical best position parameter vector, the particle swarm's global best position parameter vector, and the dynamic inertia weight coefficient to generate the updated velocity parameter vector of the particle unit can be implemented through the following example.

[0084] For each particle unit, obtain the value of each velocity component in the velocity parameter vector of the current round of that particle unit; For each particle unit, obtain the value of each position component in the individual historical best position parameter vector of that particle unit; For each particle unit, obtain the value of each position component in the global optimal position parameter vector of the particle swarm population; For each particle unit, obtain the value of each position component in the current round position parameter vector of that particle unit; Generate a first random number sequence, the length of which is equal to the number of dimensions of the control parameters, and each random number value in the first random number sequence is a random number that is uniformly distributed within a preset first random number value range; Generate a second random number sequence. The length of the second random number sequence is equal to the number of dimensions of the control parameters. Each random number value in the second random number sequence is a random number that is uniformly distributed within a preset range of second random number values. Read the preset first acceleration factor coefficient and the preset second acceleration factor coefficient; For each particle unit, the individual cognitive update component of the particle unit is calculated based on the difference between the position component value in the individual historical best position parameter vector of the particle unit and the position component value in the current round position parameter vector of the particle unit, combined with the corresponding random number value in the first random number sequence and the first acceleration factor coefficient. For each particle unit, the social learning update component of the particle unit is calculated by combining the difference between the position component value in the global optimal position parameter vector of the particle swarm population and the position component value in the current round position parameter vector of the particle unit, the corresponding random number value in the second random number sequence and the second acceleration factor coefficient. For each particle unit, the velocity component value in the current cycle velocity parameter vector of the particle unit is multiplied by the dynamic inertia weight coefficient to obtain the inertia maintenance velocity component of the particle unit. For each particle unit, the inertial maintenance velocity component, the individual cognitive update component, and the social learning update component of the particle unit are vector superimposed to generate the initial updated velocity parameter vector of the particle unit. The initial updated velocity parameter vector of each particle unit is output as the updated velocity parameter vector of that particle unit.

[0085] In this embodiment of the invention, for example, the server performs a velocity update operation for each particle unit. Taking the processing of particle #1 in the first iteration (t=1) as an example, the server first obtains its current-cycle velocity parameter vector from the particle's state structure in memory, for example (15, -0.21, 0.008, 0.33, -0.05). The server then obtains the particle's individual historical best position parameter vector, for example (228, 4.26, 0.042, 5.12, 0.79). The server also obtains the global best position parameter vector of the particle swarm, for example (190, 3.80, 0.038, 4.56, 0.82). The server also obtains the current-cycle position parameter vector of particle #1, for example (228, 4.26, 0.042, 5.12, 0.79).

[0086] Subsequently, the server generates two random number sequences. The server invokes a uniformly distributed random number generator to generate a first random number sequence of length 5 (equal to the number of dimensions in the control parameters), for example, (0.5, 0.8, 0.2, 0.6, 0.4), where each random number is within the interval [0, 1]. The server then generates a second random number sequence of the same length, for example, (0.3, 0.5, 0.9, 0.1, 0.7). The server reads the preset first acceleration factor coefficient c1=2.0 and the second acceleration factor coefficient c2=2.0 from the algorithm configuration parameters.

[0087] Next, the server calculates for each control parameter dimension separately. Taking the first dimension (dead time) as an example, the server calculates the individual cognitive update component: this component is c1*r1*(individual optimal position - current position), which is 2.0*0.5*(228-228)=0. The server calculates the social learning update component: this component is c2*r2*(global optimal position - current position), which is 2.0*0.3*(190-228)=-22.8. The server calculates the inertial maintenance velocity component: this component is w*current velocity, where the dynamic inertial weight coefficient w=0.89995, which is 0.89995*15=13.49925.

[0088] Finally, the server performs vector superposition of the three components of this dimension: Preliminary updated velocity = Inertia maintenance component + Individual cognition component + Social learning component = 13.49925 + 0 + (-22.8) = -9.30075. The server performs the above calculation process sequentially for all five dimensions to generate the preliminary updated velocity parameter vector for particle #1, for example, (-9.30075, 0.052, 0.0152, 0.254, -0.023). The server outputs this vector as the updated velocity parameter vector for this particle unit for subsequent velocity boundary constraint processing.

[0089] In this embodiment of the invention, for each particle unit, the modified position parameter vector of the particle unit is input into the energy efficiency evaluation function of the PFC system for calculation, and the current round energy efficiency evaluation value corresponding to the particle unit is used as the current round fitness value of the particle unit. This can be implemented through the following example.

[0090] For each particle unit, the value of each position component in the corrected position parameter vector of the particle unit is analyzed. Each position component value corresponds to the control parameter setting value to be executed by an adjustable execution unit in the PFC system. Generate a set of control parameter configurations required for the simulated operation of the PFC system based on the position component values ​​in the corrected position parameter vector; Call the pre-built PFC system energy efficiency simulation model, and input the control parameter configuration set and the real-time energy efficiency status parameter set into the PFC system energy efficiency simulation model; The PFC system energy efficiency simulation model is used to simulate the operating state of the PFC system under the control parameter configuration set, and a set of simulation operation results is generated. The set of simulation operation results includes a sequence of simulated output side power quality parameters and a sequence of simulated power conversion unit loss parameters. Extract the simulated values ​​of voltage harmonic compensation residual and current harmonic compensation residual from the simulated output side power quality parameter sequence from the set of simulated operation results; Extract the simulated values ​​of semiconductor device switching loss, magnetic component iron loss, and magnetic component copper loss from the simulated power conversion unit loss parameter sequence in the set of simulation results. The output-side power quality compliance index is calculated based on the simulated values ​​of the residual voltage harmonic compensation and the residual current harmonic compensation. The comprehensive loss index of the power conversion unit is calculated based on the simulated values ​​of switching losses of semiconductor devices, iron losses of magnetic components, and copper losses of magnetic components. The output-side power quality compliance index and the power conversion unit comprehensive loss index are weighted and summed to generate a comprehensive energy efficiency evaluation value. The comprehensive energy efficiency evaluation value is output as the current round fitness value of the particle unit.

[0091] In this embodiment of the invention, for example, the server performs the calculation of the particle unit fitness value. Taking particle #1 as an example, the server parses its corrected position parameter vector (218.7ns, 4.31, 0.057, 5.0, 0.77). The server maps each position component value to a set of control parameter configurations required for the simulation operation of the PFC system, such as "dead time = 218.7 nanoseconds, A-phase current loop proportional gain = 4.31, ...".

[0092] The server calls a pre-built PFC system energy efficiency simulation model (a mathematical model based on circuit topology and control principles) in memory. The server inputs the above-mentioned set of control parameters, along with the set of energy efficiency status parameters (such as current input voltage and current waveforms) obtained in real time from sensors, into the simulation model.

[0093] The simulation model begins its calculations. Based on the input control parameters and initial state, the model simulates the dynamic operation of the PFC system within one power frequency cycle. After the model completes its calculations, it returns a set of simulation results to the server. The server extracts the simulated output-side voltage harmonic residuals (e.g., 1.15%) and current harmonic residuals (e.g., 2.3%) from this set. The server also extracts the simulated power conversion unit loss sequence and summarizes the simulated values ​​for semiconductor switching losses (e.g., 19.8 watts), magnetic component iron losses (e.g., 5.1 watts), and copper losses (e.g., 3.2 watts).

[0094] The server calculates the output-side power quality compliance index (e.g., a score between 0 and 1, 0.92) based on the extracted simulated values ​​of various losses. The server also calculates the overall loss index of the power conversion unit (e.g., total loss of 28.1 watts) based on the extracted simulated values ​​of various losses. The server then weights and sums the two indices according to preset weights (e.g., power quality weight 0.4, loss weight 0.6) to generate a comprehensive energy efficiency assessment value (e.g., a percentage score of 85.7). The server outputs and records this value of 85.7 as the current round fitness value for particle #1.

[0095] In this embodiment of the invention, the dynamic inertia weight coefficient corresponding to the current iteration round is calculated based on the value of the current iteration round counter. The dynamic inertia weight coefficient decreases nonlinearly as the value of the current iteration round counter increases. This can be implemented through the following example.

[0096] Read the preset initial value parameter of inertial weight and the preset termination value parameter of inertial weight, wherein the value of the initial value parameter of inertial weight is greater than the value of the termination value parameter of inertial weight; Read the preset maximum iteration round threshold parameter to obtain the maximum number of iterations corresponding to the maximum iteration round threshold parameter; Obtain the value of the current iteration round counter as the number of iterations completed; The difference between the initial value parameter and the final value parameter of the inertial weight is calculated to obtain the total change in the inertial weight. Calculate the ratio of the current number of iterations completed to the maximum number of iterations to obtain the iteration progress ratio coefficient; The iteration progress ratio coefficient is subjected to nonlinear mapping processing to generate a nonlinear decay factor. The nonlinear decay factor changes slowly when the iteration progress ratio coefficient is small and changes quickly when the iteration progress ratio coefficient is large. Multiply the total change in inertia weight by the nonlinear decay factor to obtain the inertia weight decay amount corresponding to the current iteration round; Subtract the inertia weight decay amount corresponding to the current iteration from the initial value parameter of the inertia weight to obtain the preliminary calculated value of the dynamic inertia weight coefficient corresponding to the current iteration. Determine whether the preliminary calculated value of the dynamic inertia weight coefficient is less than the inertia weight termination value parameter; If the initial calculated value of the dynamic inertia weight coefficient is less than the inertia weight termination value parameter, then the inertia weight termination value parameter is used as the dynamic inertia weight coefficient corresponding to the current iteration round. If the initial calculated value of the dynamic inertia weight coefficient is greater than or equal to the inertia weight termination value parameter, then the initial calculated value of the dynamic inertia weight coefficient is used as the dynamic inertia weight coefficient corresponding to the current iteration round. Output the dynamic inertia weight coefficient corresponding to the current iteration round.

[0097] In this embodiment of the invention, for example, the server first reads preset parameters from the algorithm configuration file. The server reads the preset initial value parameter w_initial=0.9 for inertia weight and the preset final value parameter w_final=0.4 for inertia weight. The server also reads the preset maximum iteration threshold parameter T_max=100 and obtains its corresponding maximum iteration count value of 100.

[0098] Next, the server retrieves the current iteration state. The server accesses the current iteration round counter stored in memory and obtains its value as the current number of iterations completed, t. For example, at the start of the first iteration, t=1.

[0099] The server then begins calculations. The server calculates the total change in inertia weight: Δw = w_initial - w_final = 0.9 - 0.4 = 0.5. The server calculates the iteration progress ratio coefficient: ratio = t / T_max = 1 / 100 = 0.01.

[0100] Then, the server performs nonlinear mapping processing. Based on preset nonlinear rules (such as using a square function to accelerate decay in the later stages of iteration), the server calculates the nonlinear decay factor: decay_factor = ratio 2 =0.01 2 =0.0001. When the progress ratio is small (e.g., 0.01), the value of this factor is very small (0.0001) and changes slowly; when the progress ratio is large (e.g., 0.9), its value (0.81) is close to 1 and changes rapidly.

[0101] The server calculates the inertia weight decay for the current round: w_decay = Δw * decay_factor = 0.5 * 0.0001 = 0.00005. The server calculates the preliminary value of the dynamic inertia weight coefficient: w_temp = w_initial - w_decay = 0.9 - 0.00005 = 0.89995.

[0102] Next, the server performs boundary checks. The server checks whether w_temp(0.89995) is less than w_final(0.4). Since 0.89995 > 0.4, the condition is not met, so the server determines the value of w_temp, 0.89995, as the dynamic inertia weight coefficient w corresponding to the current iteration round.

[0103] As the iterations proceed, the server repeats this process. For example, when the 50th iteration (t=50) is reached, the server calculates ratio=50 / 100=0.5 and decay_factor=0.5. 2 =0.25, w_decay=0.5*0.25=0.125, w_temp=0.9-0.125=0.775. After the judgment, output w=0.775. When the 90th iteration (t=90) is reached, ratio=0.9, decay_factor=0.81, w_decay=0.5*0.81=0.405, w_temp=0.9-0.405=0.495, output w=0.495.

[0104] When the 100th iteration (t=100) is reached, ratio=1.0, decay_factor=1.0, w_decay=0.5, and w_temp=0.9-0.5=0.4. At this point, w_temp equals w_final, and after checking, w=0.4 is output.

[0105] Finally, the server outputs the calculation results. The server writes the calculated dynamic inertia weight coefficient w into a designated variable in memory for use in the velocity update formula of all particles in this iteration. Through this nonlinear decreasing mechanism, the algorithm assigns particles higher inertia (e.g., 0.9) in the early stages, which is beneficial for global exploration; and assigns lower inertia (e.g., 0.4) in the later stages, which is beneficial for fine-tuning near the optimal solution, thereby improving optimization efficiency and accuracy.

[0106] In this embodiment of the invention, after updating the velocity parameter vector and position parameter vector of the particle unit in each iteration based on the historical optimal position parameter vector of the particle unit in the energy efficiency optimization parameter search space of the PFC system, the global optimal position parameter vector of the particle swarm, and the dynamic inertia weight coefficient corresponding to the current iteration, the following implementation method is also provided.

[0107] After the dynamic inertia weight coefficient corresponding to the current iteration round is updated, check whether the value of the dynamic inertia weight coefficient corresponding to the current iteration round is less than the preset weight perturbation trigger threshold. If the value of the dynamic inertia weight coefficient corresponding to the current iteration round is less than the preset weight perturbation trigger threshold, then the particle swarm population diversity enhancement process will be entered. A predetermined proportion of particle units are randomly selected from the particle swarm population as the set of particle units to be disturbed. For each particle unit in the set of particles to be disturbed, a random perturbation direction vector with the same dimension as the corrected position parameter vector of the particle unit to be disturbed is generated. The value of each component in the random perturbation direction vector is a random number that is uniformly distributed within a preset perturbation amplitude range. For each particle element to be disturbed, the corrected position parameter vector of the particle element to be disturbed is added to the corresponding random vector of the disturbance direction to generate the initial disturbed position parameter vector of the particle element to be disturbed. For each position component of the initial perturbation position parameter vector of each particle unit to be perturbed, position boundary restriction processing is performed. Position components that exceed the upper limit of the corresponding control parameter dimension are restricted to the upper limit of the adjustment value, and position components that are lower than the lower limit of the corresponding control parameter dimension are restricted to the lower limit of the adjustment value, thus obtaining the perturbed position parameter vector of the particle unit to be perturbed. For each particle unit to be disturbed, the perturbed position parameter vector of the particle unit is input into the PFC system energy efficiency evaluation function for calculation to obtain the perturbed fitness value of the particle unit. For each particle unit to be perturbed, compare the perturbed fitness value of the particle unit with the individual historical best fitness value of the particle unit. If the perturbed fitness value is better than the individual historical best fitness value, then update the perturbed position parameter vector of the particle unit to the individual historical best position parameter vector of the particle unit, and update the perturbed fitness value to the individual historical best fitness value of the particle unit. Compare the fitness values ​​of all perturbed particle units with the global optimal fitness value of the particle swarm. If there is at least one perturbed particle unit whose fitness value after perturbation is better than the global optimal fitness value, then select the perturbed particle unit with the best fitness value from all perturbed particle units whose fitness value after perturbation is better than the global optimal fitness value as the perturbed global optimal particle unit. The perturbation-post-position parameter vector of the globally optimal particle unit is updated to the globally optimal position parameter vector of the particle swarm population, and the perturbation-post-fitness value of the globally optimal particle unit is updated to the globally optimal fitness value of the particle swarm population.

[0108] In this embodiment of the invention, for example, the server first detects whether a perturbation has been triggered. The server reads a preset weight perturbation trigger threshold, for example, w_trigger=0.5. The server compares the dynamic inertia weight coefficient w calculated in the current iteration (e.g., w=0.66 in the 60th iteration) with this threshold. Since 0.66>0.5, the condition is not triggered, the server skips the perturbation process and directly performs subsequent fitness evaluation and global optimum update.

[0109] As the iteration progresses to the later stages, the weights decrease, triggering a perturbation. Assuming it's the 85th iteration, the server calculates the current dynamic inertia weight coefficient w = 0.46. The server detects that 0.46 < 0.5, meeting the trigger condition, and then enters the particle swarm optimization process to enhance population diversity.

[0110] The server begins executing the perturbation operation. From the entire particle swarm (50 particles in total), the server randomly selects a preset percentage (e.g., 20%) of particles as the set of particles to be perturbed. The server invokes a random selection algorithm, possibly selecting particles #7, #12, #19, #23, #31, #34, #40, #44, #47, and #50, a total of 10 particles.

[0111] For each particle in the set to be perturbed, the server generates a perturbation vector. Taking particle #23 as an example, its current corrected position parameter vector is (202ns, 3.88, 0.040, 4.70, 0.80). The server generates a random perturbation direction vector with the same dimensions as this vector (5 dimensions). The server's preset perturbation amplitude range is [-5%, +5%] of the corresponding parameter dimension adjustment range. Taking the dead time dimension (range 400ns) as an example, 5% is 20ns. The server uniformly and randomly generates a value within this range [-20ns, +20ns], for example, +12ns. The server performs the same operation on other dimensions, generating a complete perturbation random vector, for example (+12, -0.15, +0.002, +0.10, -0.03).

[0112] The server calculates the new position after the perturbation. The server performs a vector addition operation between the current position vector of particle #23 and the perturbation random vector: (202, 3.88, 0.040, 4.70, 0.80) + (+12, -0.15, +0.002, +0.10, -0.03) = (214, 3.73, 0.042, 4.80, 0.77). This is the position parameter vector after the initial perturbation.

[0113] The server performs position boundary constraint processing. The server checks each component of the new vector. For example, the dead time of 214ns is within [100ns, 500ns], so it's valid; but is the Kp parameter 3.73 of phase A below its lower limit of 0.1? No, 3.73 is greater than 0.1, so it's valid; however, is the Kp_B parameter 4.80 above its upper limit of 5.0? No, 4.80 is less than 5.0, so it's valid. All components are within the bounds, therefore the perturbed position parameter vector for particle #23 is (214ns, 3.73, 0.042, 4.80, 0.77).

[0114] The server evaluates the fitness after the perturbation and updates the record. The server inputs the perturbed position vector of particle #23 into the PFC system energy efficiency assessment function for calculation, obtaining its perturbed fitness value, for example, 95.3%. The server compares this value with the stored historical best fitness value of particle #23 (for example, 95.1%). Since 95.3% is better, the server updates the historical best position parameter vector of particle #23 to (214, 3.73, 0.042, 4.80, 0.77) and updates its historical best fitness value to 95.3%.

[0115] After completing the "perturbation-evaluation-individual update" operation for all particles to be perturbed, the server checks whether the global optimum has been updated. The server iterates through the perturbed fitness values ​​of these 10 particles and compares them with the current global optimum fitness value of the particle swarm (e.g., 95.5%). Assuming the server finds that particle #40 has the highest perturbed fitness value of 95.8%, which is better than the current global optimum of 95.5%, the server marks particle #40 as the global optimum particle after perturbation. The server updates the global optimum position parameter vector of the particle swarm to the perturbed position vector of particle #40 and updates the global optimum fitness value to 95.8%.

[0116] At this point, the diversity enhancement process for this iteration is complete. The server then continues with the standard steps of this iteration (such as updating the iteration counter and determining the termination condition). Through this perturbation mechanism, the server actively conducts small-scale random exploration of some particles in the later stages of the algorithm, effectively improving population diversity and helping to escape possible local optima, thereby finding a globally better initial distribution set of particles.

[0117] In this embodiment of the invention, the step of generating a set of real-time energy efficiency optimization control instructions for the PFC system based on the optimal energy efficiency parameter combination of the PFC system, and sending the set of real-time energy efficiency optimization control instructions to the corresponding adjustable execution unit in the PFC system to trigger the parameter adjustment operation of the execution unit, can be implemented through the following example.

[0118] The value of each parameter component in the optimal energy efficiency parameter combination of the PFC system is analyzed, and each parameter component value corresponds to a target adjustable execution unit in the PFC system. Obtain the real-time operating parameter values ​​of each adjustable execution unit in the PFC system at the current moment; For each target adjustable execution unit, compare the parameter component values ​​corresponding to the target adjustable execution unit with the real-time operating parameter values ​​of the target adjustable execution unit, and calculate the parameter adjustment deviation between the two. For each target adjustable execution unit, the adjustment direction indicator and adjustment magnitude percentage of the target adjustable execution unit are determined according to the parameter adjustment deviation corresponding to the target adjustable execution unit; For each target adjustable execution unit, according to the device control protocol format of the target adjustable execution unit, the adjustment direction identifier and adjustment range percentage of the target adjustable execution unit are encapsulated into a unit control instruction data packet that conforms to the device control protocol format; Add timestamp and instruction priority information to each unit control instruction data packet. The timestamp is the current system time, and the instruction priority information is preset according to the importance of the corresponding adjustable execution unit in the PFC system. All unit control instruction data packets with timestamp and instruction priority information are sorted according to the instruction priority information to generate an instruction execution queue with a sequential execution order. The unit control instruction data packets are extracted sequentially from the instruction execution queue, and the extracted unit control instruction data packets are sent to the control interface of the corresponding target adjustable execution unit through the control bus inside the PFC system. After sending each unit control instruction data packet, wait for and receive the instruction execution confirmation signal returned from the corresponding target adjustable execution unit; Based on the received instruction execution confirmation signals from all target adjustable execution units, generate an execution result feedback report of the PFC system real-time energy efficiency optimization control instructions; The execution result feedback report is stored in the historical instruction execution record database of the PFC system.

[0119] In this embodiment of the invention, exemplarily, firstly, the server parses the optimal parameter combination. The server reads the optimal energy efficiency parameter combination of the PFC system stored in memory. This combination is a position parameter vector, for example (dead time = 205ns, A-phase Kp = 3.95, A-phase Ki = 0.041, B-phase Kp = 4.74, modulation ratio = 0.81). The server parses this vector and identifies that each parameter component corresponds to a target adjustable execution unit in the PFC system: dead time corresponds to the "dead time regulator", A-phase Kp corresponds to the "A-phase current loop proportional gain regulator", and so on.

[0120] Next, the server obtains the current real-time operating parameters. The server sends a query command to the central controller of the PFC system via the control bus to obtain the current real-time operating parameter values ​​for each adjustable execution unit. The server receives a response, for example: the current value of the dead time regulator is 220ns, the current value of the A-phase Kp regulator is 4.10, the current value of the A-phase Ki regulator is 0.050, the current value of the B-phase Kp regulator is 5.00, and the current value of the modulation ratio regulator is 0.78.

[0121] Then, the server calculates the adjustment deviation for each target unit. The server compares the optimal value with the current value one by one. For the dead-time regulator: deviation = optimal value 205ns - current value 220ns = -15ns. For the A-phase Kp regulator: deviation = 3.95 - 4.10 = -0.15. The server calculates the deviation for all units in sequence.

[0122] The server then determines the adjustment direction and magnitude. Based on the sign and magnitude of the deviation, the server determines the adjustment direction indicator ("increase" or "decrease") and the adjustment magnitude percentage. Taking the dead-time regulator as an example, a deviation of -15ns is negative, so the direction indicator is "decrease". The magnitude percentage = |deviation| / adjustment range * 100%. The dead-time adjustment range is 400ns, so the magnitude percentage = 15 / 400 * 100% = 3.75%. The server calculates similar results for each unit.

[0123] Next, the server encapsulates the unit control instruction data packet. The server encapsulates the data according to the device control protocol format of each adjustable execution unit. For example, the "dead-time regulator" uses the Modbus RTU protocol. According to its protocol format, the server fills the data frame with the register address (e.g., 0x1001), function code (e.g., 0x06 write a single register), and data containing the "reduce by 3.75%" instruction encoding (e.g., converting 205ns to the corresponding register value), generating a complete unit control instruction data packet. For the "A-phase Kp regulator" using a different protocol (e.g., CAN protocol), the server encapsulates another data packet according to the CAN frame format.

[0124] The server adds tags and sorts the instruction packets. For each generated unit control instruction packet, the server adds a timestamp (e.g., 2023-10-27 14:30:05.123) and an instruction priority tag. Priorities are set according to preset rules; for example, the "modulation ratio regulator" instruction, which directly affects system stability, has the highest priority, while the "dead time regulator" has a high priority, and the "PI parameter regulator" has a medium priority. The server sorts all tagged instruction packets according to priority from highest to lowest, generating an instruction execution queue. In this example, the "modulation ratio regulator" instruction packet will be at the front of the queue.

[0125] Next, the server sends instructions sequentially and waits for confirmation. Starting from the head of the queue, the server retrieves unit control instruction data packets one by one. First, the server sends the Modbus TCP instruction packet for the "Modulation Ratio Regulator" to the corresponding industrial controller network address via Ethernet. After sending, the server starts a timeout timer (e.g., 2 seconds) and listens for the TCP acknowledgment message returned by the controller. Upon receiving a correct acknowledgment message for the "Modulation Ratio Regulator," the server then retrieves the next instruction packet for the "Dead Time Regulator" from the queue, sends it via the RS-485 bus, and waits for its Modbus RTU format acknowledgment frame. The server processes all instruction packets in the queue sequentially.

[0126] Finally, the server generates and stores a feedback report. After all instructions have been sent or timed out, the server summarizes the received confirmation signals. Assume that all five target execution units returned confirmation signals before the timeout. The server generates a "PFC System Real-Time Energy Efficiency Optimization Control Instruction Execution Result Feedback Report." The report includes: optimization task ID, issuance timestamp, instruction content for each target unit, expected value, sending status (success / timeout), confirmation status (confirmed / unconfirmed), and actual completion timestamp. The server writes this structured report to a database table named "PFC_Control_History" through a database interface, completing the full closed loop of this real-time energy efficiency optimization control.

[0127] This invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned real-time energy efficiency optimization method for a PFC system based on a dynamic weighted particle swarm optimization algorithm. Figure 2 As shown, Figure 2 This is a structural block diagram of a computer device 100 provided in an embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112, and a communication unit 113. To enable data transmission or interaction, the memory 111, processor 112, and communication unit 113 are electrically connected to each other directly or indirectly. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0128] For illustrative purposes, the foregoing description has been made with reference to specific embodiments. However, the foregoing illustrative discussions are not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed. Numerous modifications and variations are possible in accordance with the foregoing teachings. These embodiments were chosen and described in order to best illustrate the principles of the present disclosure and its practical application, thereby enabling those skilled in the art to best utilize the disclosure and to employ various embodiments with different modifications to suit a particular intended application.

Claims

1. A real-time energy efficiency optimization method for a PFC system based on dynamic weighted particle swarm optimization algorithm, characterized in that, include: Constructing a search space for energy efficiency optimization parameters of PFC systems; Obtain the set of real-time energy efficiency status parameters of the PFC system under its current operating conditions; The particle swarm population is initialized based on the search space of the PFC system energy efficiency optimization parameters and the set of real-time energy efficiency status parameters to obtain an initial distribution set of the particle swarm. The initial distribution set of the particle swarm contains multiple particle units. Each particle unit has an initial position parameter vector and an initial velocity parameter vector in the search space of the PFC system energy efficiency optimization parameters. The initial position parameter vector represents a set of control parameters to be executed by the PFC system. The dynamic weighted particle swarm optimization algorithm is used to perform multiple rounds of iterative optimization on the initial distribution set of the particle swarm. By updating the particle velocity and position based on historical optimal, global optimal and dynamic inertia weight, the global optimal parameters are output as the optimal energy efficiency parameter combination of the PFC system. A set of real-time energy efficiency optimization control instructions for the PFC system is generated based on the optimal energy efficiency parameter combination of the PFC system, and the set of real-time energy efficiency optimization control instructions for the PFC system is sent to the corresponding adjustable execution unit in the PFC system to trigger the parameter adjustment operation of the execution unit.

2. The method according to claim 1, characterized in that, The method employs a dynamic weighted particle swarm optimization algorithm to iteratively optimize the initial distribution set of the particle swarm in multiple rounds. By updating the particle velocity and position based on historical optimal, global optimal, and dynamic inertia weights, it outputs globally optimal parameters as the optimal energy efficiency parameter combination for the PFC system, including: The dynamic weighted particle swarm optimization algorithm is invoked to perform multiple rounds of iterative optimization on the initial distribution set of the particle swarm. In each iteration, the velocity parameter vector and position parameter vector of the particle unit are updated according to the historical optimal position parameter vector of the particle unit in the search space of the energy efficiency optimization parameters of the PFC system, the global optimal position parameter vector of the particle swarm population, and the dynamic inertia weight coefficient corresponding to the current iteration round. When the preset iteration termination condition is reached, the global optimal position parameter vector of the particle swarm population is output as the optimal energy efficiency parameter combination of the PFC system.

3. The method according to claim 2, characterized in that, The dynamic weighted particle swarm optimization algorithm is invoked to perform multiple rounds of iterative optimization on the initial distribution set of the particle swarm. In each iteration, the velocity and position parameter vectors of the particle units are updated based on the historical optimal position parameter vector of the particle unit in the PFC system energy efficiency optimization parameter search space, the globally optimal position parameter vector of the particle swarm population, and the dynamic inertia weight coefficient corresponding to the current iteration round. Upon reaching a preset iteration termination condition, the globally optimal position parameter vector of the particle swarm population is output as the optimal energy efficiency parameter combination for the PFC system, including: Set the current iteration round counter and initialize the value of the current iteration round counter to the initial round number value; Read the preset maximum iteration round threshold parameter as part of the iteration termination condition; Extract the current round position parameter vector and the current round velocity parameter vector of all particle units from the initial distribution set of the particle swarm; The dynamic inertia weight coefficient corresponding to the current iteration round is calculated based on the value of the counter for the current iteration round. The dynamic inertia weight coefficient decreases nonlinearly as the value of the counter for the current iteration round increases. For each particle unit, a velocity update operation is performed based on the current round velocity parameter vector of the particle unit, the individual historical best position parameter vector of the particle unit, the global best position parameter vector of the particle swarm, and the dynamic inertia weight coefficient to generate the updated velocity parameter vector of the particle unit. For each velocity component of the updated velocity parameter vector of each particle unit, velocity boundary constraint processing is performed. Velocity components exceeding the preset upper velocity limit are constrained to the upper velocity limit, and velocity components below the preset lower velocity limit are constrained to the lower velocity limit, thus obtaining the corrected velocity parameter vector of the particle unit. For each particle unit, a position update operation is performed based on the current round position parameter vector and the corrected velocity parameter vector of the particle unit to generate the updated position parameter vector of the particle unit. For each position component of the updated position parameter vector of each particle unit, position boundary restriction processing is performed. Position components that exceed the upper limit of the corresponding control parameter dimension are restricted to the upper limit of the adjustment value, and position components that are lower than the lower limit of the corresponding control parameter dimension are restricted to the lower limit of the adjustment value, thus obtaining the corrected position parameter vector of the particle unit. For each particle unit, the corrected position parameter vector of the particle unit is input into the energy efficiency evaluation function of the PFC system for calculation, and the current round energy efficiency evaluation value of the particle unit is obtained as the current round fitness value of the particle unit. For each particle unit, compare the current fitness value of the particle unit with the individual historical best fitness value of the particle unit. If the current fitness value is better than the individual historical best fitness value, then update the corrected position parameter vector of the particle unit to the individual historical best position parameter vector of the particle unit, and update the current fitness value of the particle unit to the individual historical best fitness value of the particle unit. By comparing particle fitness values ​​and updating the global optimal solution during particle swarm optimization iterations, the optimal energy efficiency parameter combination of the PFC system is output after reaching the maximum number of iterations.

4. The method according to claim 3, characterized in that, The process of comparing particle fitness values ​​and updating the global optimal solution during particle swarm optimization iterations until the maximum number of iterations is reached, and then outputting the optimal energy efficiency parameter combination of the PFC system, includes: Compare the current fitness value of all particle units with the global optimal fitness value of the particle swarm. If there is at least one particle unit whose current fitness value is better than the global optimal fitness value, then select the particle unit with the best fitness value from all particle units whose current fitness value is better than the global optimal fitness value as the global optimal particle unit for this round. The corrected position parameter vector of the global best particle unit in this round is updated to the global best position parameter vector of the particle swarm population, and the current round fitness value of the global best particle unit in this round is updated to the global best fitness value of the particle swarm population. The value of the current iteration round counter is increased by one iteration step unit to obtain the updated value of the current iteration round counter; Determine whether the updated value of the current iteration round counter has reached the maximum iteration round threshold parameter. If it has not reached the maximum iteration round threshold parameter, return to the step of calculating the dynamic inertia weight coefficient corresponding to the current iteration round based on the value of the current iteration round counter and continue to perform the next round of iteration optimization. If the updated current iteration round counter value reaches the maximum iteration round threshold parameter, the iterative optimization process is terminated, and the final global optimal position parameter vector is extracted from the current state of the particle swarm population. The final global optimal position parameter vector is output as the optimal energy efficiency parameter combination of the PFC system.

5. The method according to claim 1, characterized in that, The energy efficiency optimization parameter search space of the PFC system consists of multiple control parameter dimensions that can be dynamically adjusted during the operation of the PFC system. Each control parameter dimension corresponds to the adjustment range of the adjustable execution unit in the PFC system. The construction of the PFC system energy efficiency optimization parameter search space includes: By collecting the specifications and linkage constraints of the adjustable execution unit of the PFC system, a multidimensional coupled parameter search space that satisfies the constraints is constructed and discretized into a mesh, thus obtaining the discretized parameter search mesh space. Extract the node coordinate vector of each grid node in the discretized parameter search grid space. The node coordinate vector is composed of the center value of the sub-interval cell in each control parameter dimension. An index mapping table for the search space of energy efficiency optimization parameters of the PFC system is constructed based on the node coordinate vectors, and each node coordinate vector in the index mapping table is associated with a unique spatial index identifier; The spatial index identifiers corresponding to the coordinate vectors of all nodes in the index mapping table are combined to generate a discretized search space topology for particle swarm optimization. In the discretized search space topology, the adjacency relationships between each grid node are marked, including adjacency relationships in the same dimension and diagonal adjacency relationships between different dimensions; A grid node connectivity matrix is ​​generated based on the adjacency relationship. The row and column indices of the grid node connectivity matrix are both spatial index identifiers, and the matrix element values ​​are used to identify whether there is a direct connection path between two grid nodes in the corresponding row and column.

6. The method according to claim 5, characterized in that, The process involves collecting the specifications and linkage constraints of the adjustable execution units of the PFC system, constructing a multidimensional coupled parameter search space that satisfies the constraints, and then discretizing the mesh to obtain the discretized parameter search mesh space, including: Collect the set of device specification parameters for all adjustable execution units with dynamically adjustable parameters in the PFC system. The set of device specification parameters includes the lower limit and upper limit of adjustment for each adjustable execution unit. Each control parameter dimension is defined with respect to the lower and upper limits of the adjustable execution unit. The starting value of the dimension boundary interval is the lower limit of the corresponding adjustable execution unit, and the ending value is the upper limit of the corresponding adjustable execution unit. Extract the set of parameter linkage constraint relationships between multiple adjustable execution units with dynamically adjustable parameters in the PFC system. The set of parameter linkage constraint relationships includes parameter adjustment ratio constraint parameters and parameter adjustment sequence constraint parameters of at least two adjustable execution units. The dimensional boundary intervals are modified according to the set of parameter linkage constraints to generate a multidimensional coupled parameter search space that satisfies the parameter linkage constraints. The multidimensional coupled parameter search space has coupling and correlation edges between different dimensional boundary intervals determined based on the parameter linkage constraints. The multidimensional coupling parameter search space is subjected to spatial discretization grid partitioning. According to the preset grid partitioning density parameter, each dimension boundary interval is divided into multiple continuous sub-interval units to generate a discretized parameter search grid space with grid node coordinates.

7. The method according to claim 1, characterized in that, The real-time energy efficiency status parameter set includes power quality monitoring parameters at the input of the PFC system, power quality compensation parameters at the output of the PFC system, and loss status parameters of the power conversion unit inside the PFC system. The acquisition of the real-time energy efficiency status parameter set under the current operating conditions of the PFC system includes: The instantaneous waveform sequences of the input voltage and the input current are acquired from the power quality monitoring sensor array at the input end of the PFC system. The waveform feature extraction process is performed on the instantaneous waveform sequence of the input voltage to obtain the effective value parameter of the input voltage, the harmonic distortion rate parameter of the input voltage, and the frequency fluctuation parameter of the input voltage, which are used as components of the power quality monitoring parameters. The waveform feature extraction process is performed on the instantaneous waveform sequence of the input current to obtain the effective value parameter of the input current, the harmonic distortion rate parameter of the input current, and the phase difference parameter between the input current and the input voltage, which are used as components of the power quality monitoring parameters. The instantaneous waveform sequences of the voltage and current after compensation on the output side are collected from the power quality compensation sensor array at the output end of the PFC system. The compensation effect of the output-side compensated instantaneous voltage waveform sequence is evaluated to obtain the output-side voltage harmonic compensation residual parameter and the output-side voltage amplitude stability parameter as components of the power quality compensation parameter. The compensation effect of the instantaneous waveform sequence of the current after compensation on the output side is evaluated to obtain the residual parameters of the current harmonic compensation on the output side and the compliance parameters of the power factor compensation on the output side as components of the power quality compensation parameters. The instantaneous temperature sequences of the semiconductor device junctions and the magnetic components of the power conversion unit are collected from the temperature monitoring sensor array inside the PFC system. The instantaneous sequence of DC bus voltage and the instantaneous sequence of voltage across the switching transistors inside the power conversion unit are collected from the voltage monitoring point of the power conversion unit inside the PFC system. The instantaneous sequence of the conduction current of the switching transistors and the instantaneous sequence of the inductor current inside the power conversion unit are collected from the current monitoring point of the power conversion unit inside the PFC system. The instantaneous value of the semiconductor device switching loss of the power conversion unit is calculated based on the instantaneous sequence of the junction temperature of the semiconductor device, the instantaneous sequence of the voltage across the switch, and the instantaneous sequence of the conduction current of the switch, and is used as a component of the loss state parameter. The instantaneous values ​​of iron loss and copper loss of the magnetic components of the power conversion unit are calculated based on the instantaneous temperature sequence, inductor current sequence, and DC bus voltage sequence, and are used as components of the loss state parameters.

8. The method according to claim 5, characterized in that, The initial particle swarm population is initialized based on the search space of the PFC system energy efficiency optimization parameters and the set of real-time energy efficiency state parameters to obtain the initial distribution set of the particle swarm, including: Analyze the set of dimensional boundary intervals in the search space of the energy efficiency optimization parameters of the PFC system, and obtain the lower limit value and upper limit value of adjustment corresponding to each dimension of the control parameter; Generate multiple uniformly distributed random number sequences with the same number of particle units based on the preset particle population size parameter. The length of each uniformly distributed random number sequence is equal to the number of dimensions of the control parameter. For each particle unit, each random value in the uniformly distributed random number sequence corresponding to the particle unit is mapped to the range formed by the lower limit and upper limit of the corresponding control parameter dimension, thereby generating the initial position parameter vector of the particle unit in the energy efficiency optimization parameter search space of the PFC system. For each particle unit, a sequence of random velocity components with the same dimension as the initial position parameter vector of the particle unit is generated according to a preset initial velocity range parameter. The value of each velocity component in the random velocity component sequence is within the range of the preset initial velocity range parameter. Use the random number sequence of velocity components of each particle unit as the initial velocity parameter vector of that particle unit; For each particle unit, the initial position parameter vector of the particle unit is input into the energy efficiency evaluation function of the PFC system for calculation, and the initial energy efficiency evaluation value corresponding to the particle unit is used as the initial fitness value of the particle unit. The initial position parameter vector of each particle unit is used as the individual historical best position parameter vector of that particle unit, and the initial fitness value of each particle unit is used as the individual historical best fitness value of that particle unit. Compare the initial fitness values ​​of all particle units and select the particle unit with the best fitness value as the initial global optimal particle unit. The initial position parameter vector of the initial globally optimal particle unit is used as the globally optimal position parameter vector of the particle swarm population, and the initial fitness value of the initial globally optimal particle unit is used as the globally optimal fitness value of the particle swarm population. A particle unit state structure is constructed based on the initial position parameter vector and initial velocity parameter vector of each particle unit. The particle unit state structure includes a particle unit identifier, a particle unit position parameter vector in the current round, a particle unit velocity parameter vector in the current round, a particle unit's historical best position parameter vector, and a particle unit's historical best fitness value. Aggregate the particle unit state structures corresponding to all particle units to generate an initial distribution set of the particle swarm containing complete particle unit state information.

9. The method according to claim 1, characterized in that, The step of generating a set of real-time energy efficiency optimization control instructions for the PFC system based on the optimal energy efficiency parameter combination of the PFC system, and sending the set of real-time energy efficiency optimization control instructions to the corresponding adjustable execution unit in the PFC system to trigger the parameter adjustment operation of the execution unit includes: The value of each parameter component in the optimal energy efficiency parameter combination of the PFC system is analyzed, and each parameter component value corresponds to a target adjustable execution unit in the PFC system. Obtain the real-time operating parameter values ​​of each adjustable execution unit in the PFC system at the current moment; For each target adjustable execution unit, compare the parameter component values ​​corresponding to the target adjustable execution unit with the real-time operating parameter values ​​of the target adjustable execution unit, and calculate the parameter adjustment deviation between the two. For each target adjustable execution unit, the adjustment direction indicator and adjustment magnitude percentage of the target adjustable execution unit are determined according to the parameter adjustment deviation corresponding to the target adjustable execution unit; For each target adjustable execution unit, according to the device control protocol format of the target adjustable execution unit, the adjustment direction identifier and adjustment range percentage of the target adjustable execution unit are encapsulated into a unit control instruction data packet that conforms to the device control protocol format; Add timestamp and instruction priority information to each unit control instruction data packet. The timestamp is the current system time, and the instruction priority information is preset according to the importance of the corresponding adjustable execution unit in the PFC system. All unit control instruction data packets with timestamp and instruction priority information are sorted according to the instruction priority information to generate an instruction execution queue with a sequential execution order. The unit control instruction data packets are extracted sequentially from the instruction execution queue, and the extracted unit control instruction data packets are sent to the control interface of the corresponding target adjustable execution unit through the control bus inside the PFC system. After sending each unit control instruction data packet, wait for and receive the instruction execution confirmation signal returned from the corresponding target adjustable execution unit; Based on the received instruction execution confirmation signals from all target adjustable execution units, generate an execution result feedback report of the PFC system real-time energy efficiency optimization control instructions; The execution result feedback report is stored in the historical instruction execution record database of the PFC system.

10. A server system, characterized in that, Includes a server for performing the method according to any one of claims 1-9.

Images