Electric energy quality compensation control method and system based on neural network

Abstract

The invention discloses an electric energy quality compensation control method and system based on a neural network, and the method comprises the steps: converting a three-phase voltage into a complex space vector to retain amplitude and phase information, employing a constructed complex reference vector as a reference, and carrying out the self-adaptive iteration updating of a weight according to the deviation between an expected signal and a current state. According to the method, a mechanism capable of automatically adjusting the learning ability according to power grid working condition changes is established, rapid convergence can be performed to track large signal disturbance when power grid voltage sags or suddenly changes, weight stability can be maintained to suppress high-frequency noise during steady-state operation, and therefore the restrictive relation between the response speed and compensation precision is effectively broken through; and a negative sequence compensation instruction is accurately generated, and efficient and flexible treatment of the power quality problem of the power distribution network is realized.

Description

Technical Field

[0001] This application relates to the field of intelligent control, and more specifically, to a power quality compensation control method and system based on neural networks. Background Technology

[0002] With the large-scale integration of new energy sources such as distributed photovoltaic power and the increasing complexity of load characteristics, distribution networks face severe challenges from various power quality issues, including voltage fluctuations, three-phase imbalance, and harmonic pollution. Constructing an efficient and precise power quality compensation and control scheme is of paramount importance for ensuring power supply reliability and improving the economic efficiency and security of power grid operation.

[0003] Currently, controllers based on adaptive linear neural networks (ADALINE) have become an advanced solution in the field of power quality compensation. They replace traditional PI regulators with online learned weights, enabling rapid tracking and compensation of harmonics, reactive power, and imbalance components. However, these solutions typically rely on least mean square (LMS) algorithms for weight updates, and their inherent fixed step-size mechanism has significant limitations when dealing with dynamic disturbances in the power grid. Specifically, during transient events such as voltage sags, a larger step size is needed for rapid response, but this leads to increased harmonic content and decreased accuracy during steady-state operation. Conversely, using a small step size to ensure steady-state accuracy results in slow dynamic convergence, making it difficult to meet the rapid compensation requirements in the early stages of a fault.

[0004] Therefore, an optimized neural network-based power quality compensation control method is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application provides a power quality compensation control method and system based on neural networks.

[0006] According to one aspect of this application, a power quality compensation control method based on a neural network is provided, comprising: Obtain the instantaneous values ​​of the three-phase voltage; The instantaneous values ​​of the three-phase voltages are subjected to complex space vector transformation to obtain complex voltage space vectors, and the complex voltage space vectors are designated as the desired complex signals; A complex reference vector is constructed based on the fundamental angular frequency of the power grid and the sampling period. Based on the complex reference vector and the desired complex signal, the complex weight vector is adaptively updated to obtain the updated complex weight vector; Based on the updated complex weight vector, generate negative order compensation complex instructions.

[0007] According to another aspect of this application, a power quality compensation control system based on a neural network is provided, comprising: Voltage acquisition module, used to acquire instantaneous three-phase voltage values; The complex space vector transformation module is used to perform complex space vector transformation on the instantaneous values ​​of three-phase voltages to obtain complex voltage space vectors, and to designate the complex voltage space vectors as the desired complex signals; The complex reference construction module is used to construct a complex reference vector based on the fundamental angular frequency of the power grid and the sampling period; The complex weight update module is used to adaptively update the complex weight vector based on the complex reference vector and the desired complex signal to obtain the updated complex weight vector. The negative-order compensation complex number instruction generation module is used to generate negative-order compensation complex number instructions based on the updated complex number weight vector.

[0008] Compared with existing technologies, this application provides a power quality compensation control method and system based on neural networks. This method transforms three-phase voltages into complex spatial vectors to retain amplitude and phase information, uses a constructed complex reference vector as a reference, and adaptively iteratively updates weights based on the deviation between the desired signal and the current state. This method establishes a mechanism that can automatically adjust its learning ability according to changes in grid operating conditions. It can quickly converge to track large signal disturbances during grid voltage dips or sudden changes, while maintaining weight stability to suppress high-frequency noise during steady-state operation. This effectively breaks the constraint between response speed and compensation accuracy, accurately generates negative-sequence compensation commands, and achieves efficient and flexible management of power quality problems in distribution networks. Attached Figure Description

[0009] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 This is a flowchart of a neural network-based power quality compensation control method according to an embodiment of this application; Figure 2 This is a schematic diagram of the data flow of a neural network-based power quality compensation control method according to an embodiment of this application; Figure 3 This is a block diagram of a neural network-based power quality compensation control system according to an embodiment of this application. Detailed Implementation

[0011] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0012] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0013] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0014] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0015] The technical solution of this application proposes a power quality compensation control method based on neural networks. Figure 1 This is a flowchart of a neural network-based power quality compensation control method according to an embodiment of this application. Figure 2 This is a system architecture diagram of a neural network-based power quality compensation control method according to an embodiment of this application. Figure 1 and Figure 2 As shown, the power quality compensation control method based on neural networks according to an embodiment of this application includes the following steps: S1, obtaining the instantaneous values ​​of three-phase voltages; S2, performing a complex space vector transformation on the instantaneous values ​​of three-phase voltages to obtain a complex voltage space vector, and designating the complex voltage space vector as the desired complex signal; S3, constructing a complex reference vector based on the fundamental angular frequency of the power grid and the sampling period; S4, performing adaptive weighting updates on the complex weight vector based on the complex reference vector and the desired complex signal to obtain an updated complex weight vector; S5, generating a negative-order compensation complex command based on the updated complex weight vector.

[0016] Specifically, S1 involves acquiring the instantaneous values ​​of the three-phase voltage. Since the distribution network faces severe challenges from various power quality issues such as voltage fluctuations, three-phase imbalance, and harmonic pollution, and the UPQC-based flexible power quality control device aims to "unifiedly solve distribution network operation problems caused by load power consumption and distributed photovoltaic access, such as three-phase imbalance, reactive power backflow, harmonic pollution, and load voltage fluctuations and sags / swells," the first step is to acquire real-time physical quantity information of the power grid. Specifically, in order for the device to monitor distribution network power quality parameters (such as voltage sags / swells, voltage harmonics, voltage imbalance, current harmonics, power factor, etc.) in real time as power quality compensation targets, the system needs to collect grid voltage signals at high frequency. This provides the necessary input data for the subsequent control system to use the ADLINE neural network algorithm as the core voltage and current loop control algorithm, ensuring faster and more accurate tracking and control of the compensation target.

[0017] The instantaneous three-phase voltage values ​​refer to the actual amplitude values ​​of the three-phase voltages (A, B, and C) in the distribution network at discrete sampling points. This differs from the effective value (RMS). The instantaneous values ​​fully preserve the phase, frequency, and distortion information of the voltage waveform, serving as the original data basis for time-domain analysis and space vector transformation.

[0018] In practice, firstly, voltage sensors (such as voltage transformers (PTs) or voltage dividers) installed at the grid connection point or the bus to be compensated of the UPQC device are used to collect the three-phase (phase A, phase B, and phase C) analog signals of voltage to ground or line voltage in real time. These sensors proportionally convert the high-voltage signals into low-voltage signals that can be processed by subsequent circuits. Subsequently, these low-voltage analog signals are sent to a signal conditioning circuit for necessary filtering (to eliminate high-frequency noise interference), amplification or attenuation (to adjust the signal amplitude to a suitable input range for the analog-to-digital converter), and isolation protection. The conditioned three-phase analog voltage signals are then synchronously fed into a multi-channel analog-to-digital converter (ADC). The ADC samples and quantizes the analog signals at a fixed, sufficiently high sampling frequency (determined by the sampling period in the control system, typically much higher than twice the fundamental frequency of the grid to satisfy the Nyquist sampling theorem, for example, thousands to tens of thousands of times per second), converting continuous analog voltage values ​​into discrete digital sequences. This discrete digital sequence is the instantaneous value of the three-phase voltage. This process ensures accurate and reliable conversion between physical grid signals and digital quantities that can be processed by digital controllers (such as DSPs and FPGAs).

[0019] Specifically, in step S2, a complex space vector transformation is performed on the instantaneous values ​​of the three-phase voltages to obtain a complex voltage space vector, and this complex voltage space vector is designated as the desired complex signal. It should be understood that directly performing real-time decomposition and extraction calculations of voltage harmonics, imbalance, etc., in the three-phase abc coordinate system is extremely complex. Therefore, in the technical solution of this application, a complex space vector transformation (i.e., Clarke transform and its generalization) is used to map the three-phase voltage signal to a two-phase orthogonal α-β coordinate system, and further express it in complex form. In this way, the positive-sequence, negative-sequence, and zero-sequence components, as well as each harmonic component, in the three-phase voltage can be effectively separated in the complex frequency domain. In particular, the positive-sequence and negative-sequence components are represented as vectors with opposite rotational directions in the complex space. This separation characteristic allows subsequent neural network algorithms (such as ADLINE networks) to achieve rapid tracking and extraction of specific components through simple complex weight updates. Furthermore, the transformed complex voltage space vector is designated as the desired complex signal, providing a clear and dynamically updated reference signal for the neural network controller. This enables the neural network to learn and calculate the amount of compensation needed to eliminate power quality problems (such as negative sequence components and harmonic components) based on the current actual grid voltage state.

[0020] In practice, a fixed transformation matrix can be applied to map the three-phase instantaneous value vectors into a complex number. This calculation process is typically implemented in a digital signal processor (DSP) or field-programmable gate array (FPGA) using software programming or hardware logic. Specifically, firstly, a complex space vector transformation is performed on the three-phase voltage instantaneous values ​​to obtain a complex voltage space vector. This process is expressed by the following formula:

[0021] in, These are the instantaneous values ​​of the three-phase voltage. Let be a complex voltage space vector. In this formula, firstly, the instantaneous values ​​of the three-phase voltages are transformed from the abc coordinate system to the two-phase stationary coordinate system using the standard Clarke transform. Using a coordinate system, we obtain two orthogonal components. and Furthermore, these two real components are combined into a complex number. This complex number is the complex voltage space vector. It is worth noting that the complex voltage space vector is a vector whose real part is... The imaginary part is The complex number of the vector, in the complex plane, has a length and rotational angular velocity that are directly related to the amplitude and frequency of the fundamental voltage of the power grid.

[0022] Secondly, the complex voltage space vector is designated as the desired complex signal. In the program implementation, a variable is defined to represent the desired complex signal, and the value of the complex voltage space vector is assigned to the desired complex signal. Through this operation, the complex representation of the grid voltage at the current moment is established as the desired input signal for the entire adaptive control system.

[0023] Specifically, in step S3, a complex reference vector is constructed based on the fundamental angular frequency of the power grid and the sampling period. It should be understood that the control system of this application employs the ADLINE neural network algorithm as the core voltage-current loop control algorithm. The core principle of ADLINE or similar linear neural networks and adaptive filtering algorithms is to approximate a desired signal (i.e., the desired complex signal obtained in the previous step) by linearly combining the input signal (i.e., the complex reference vector constructed here) with a set of weight coefficients. In order to accurately extract a specific component (e.g., the fundamental negative-sequence component) from the desired complex signal (which contains a mixture of fundamental, harmonic, positive-sequence, and negative-sequence components), a probe that changes synchronously with this specific component, i.e., the complex reference vector, needs to be provided to the neural network. Specifically, the positive-sequence and negative-sequence components in the power grid can be regarded in complex space as vectors rotating forward and backward at the fundamental angular frequency, respectively. Therefore, by constructing a complex reference vector containing forward and reverse rotation factors, two independent reference channels can be provided to the neural network for matching and extracting the positive-sequence and negative-sequence components, respectively. Specifically, a complex reference signal that rotates synchronously with the fundamental frequency of the power grid is created. This signal serves as the benchmark for updating the weights of the neural network, enabling the network to accurately separate specific components such as positive and negative sequences from the complex power grid voltage signal. The construction of this benchmark vector directly determines the accuracy and efficiency of the subsequent adaptive weight update process, and is the foundation for achieving fast and accurate compensation.

[0024] In practice, the fundamental angular frequency of the power grid and the sampling period can be converted into a complex rotation factor using Euler's formula. This calculation is performed once per sampling period in a digital signal processor (DSP) or field-programmable gate array (FPGA). Specifically, at each discrete sampling time k, two pre-set system parameters or those acquired in real time via a phase-locked loop (PLL) are used: the fundamental angular frequency of the power grid... (Unit: radians / second) and the system's sampling period Then, based on these two parameters, a two-dimensional complex reference vector is directly calculated. This process can be expressed by the following formula:

[0025] in, The sampling period represents the time interval between two consecutive samples taken by the analog-to-digital converter (ADC), and determines the time resolution of the discrete-time system. The fundamental angular frequency of the power grid. As a positive rotation factor, it represents a factor with an angular velocity in the complex plane. A unit complex vector rotating counterclockwise; The inverse rotation factor represents a factor with an angular velocity in the complex plane. A unit complex vector rotating clockwise; This is a discrete-time index that increments with each sampling period; This represents the total time elapsed from the initial moment to the current moment. This formula shows that the complex reference vector... It is a column vector containing two elements, each of which is a complex number, and their phases change continuously with time k, thus simulating the rotational characteristics of the positive and negative order components. In actual programs, this can be implemented using a lookup table or by directly calling the complex exponent function library.

[0026] Specifically, in step S4, based on the complex reference vector and the desired complex signal, the complex weight vector is adaptively updated to obtain the updated complex weight vector. It should be understood that while the traditional Complex Least Mean Square (CLMS) algorithm can eventually converge when dealing with grid imbalance problems, it suffers from a fundamental technical reason regarding dynamic response. Specifically, during weight updates, the error signal upon which this algorithm relies is a mixture of positive and negative sequence errors. This leads to an inevitable interference between the positive sequence weight update and the negative sequence estimation error during weight iteration, creating a transient coupling. The essence of this coupling phenomenon lies in the algorithm's failure to fully utilize the mathematically orthogonal and separable properties of the positive and negative sequence components to actively decouple them within a single calculation cycle. Instead, it relies on the statistical averaging effect over multiple cycles to eliminate interference. Therefore, when the grid imbalance changes drastically, this inherent coupling slows down the convergence of the two weights to their true values ​​and may even cause oscillations, thus limiting the compensation system's ability to respond to dynamic disturbances. To address the aforementioned technical deficiencies, this application proposes an adaptive update mechanism based on sequential error feedforward decoupling. By reconstructing the parallel coupled update process into a sequentially decoupled flow, it achieves efficient utilization of information and step-by-step purification of error sources within a sampling period.

[0027] In practice, firstly, based on the expected complex signal and the complex weight vector, the positive-sequence component error is predicted and the weights are initially updated on the complex reference vector to obtain temporary updated values ​​for the positive-sequence weights. That is, when updating the weights used to estimate the positive-sequence component of the grid voltage, the interference caused by the estimation error of the negative-sequence component is eliminated in advance, thereby creating a relatively clean learning environment for the update of the positive-sequence weights, which significantly improves its convergence speed and accuracy.

[0028] In this process, firstly, a decoupled target signal specifically for the positive-sequence channel is constructed by extracting the negative-sequence component estimated by the negative-sequence weight from the original complex signal. This process is expressed by the following formula:

[0029] in, The target signal for decoupling the forward-sequence channel. For the desired complex signal, This represents the negative weight at the current moment (current negative weight). Let be the complex reference vector representing the negative order (reverse rotation factor). This calculation process proactively uses the current best estimate of the negative order component to purify the positive order channel, ensuring that the subsequently calculated positive order error more accurately reflects the estimation bias of the positive order component. Secondly, based on this target signal, the decoupling error of the positive order channel is calculated. Then, according to the Least Mean Square (LMS) algorithm criterion, along the direction of the fastest error decrease, using the positive order learning rate... The step size is the current positive order weight. The update process can be represented by the following formula:

[0030] in, This is a temporary update value for the positive order weights. The current positive order weights, For positive learning rate, The complex reference vector representing the orthogonal order (forward rotation factor) is used. This is the complex conjugate of the positive-order channel decoupling error. By isolating the main error sources, the accuracy and convergence speed of the positive-order weight update are significantly improved.

[0031] Next, based on the expected complex signal and the temporary update value of the positive-order weights, the current negative-order weights are precisely estimated for the negative-order component error and the weights are finally updated to obtain the final updated value of the negative-order weights. That is, after completing the initial update of the positive-order weights, this updated, more accurate information is used to further improve the accuracy of the negative-order weight update, thereby completely resolving the coupling problem in the positive-order and negative-order weight update processes. Specifically, when performing the negative-order weight update, the latest information—the temporary update value of the positive-order weights just calculated in the current period—is actively and feedforwardly used to subtract the positive-order component from the expected signal. This provides a deeply purified target signal for the negative-order weight update calculation, in which the error of the positive-order component has been significantly reduced. This allows the negative-order weight update to be based on a better understanding of the system state, ultimately achieving rapid convergence and high accuracy in the estimation of the negative-order component.

[0032] In this process, firstly, the negative-order component error is accurately estimated for the current negative-order weights to construct a deeply decoupled target signal for the negative-order channel. Specifically, this is achieved from the currently sampled desired complex signal. In the next step, the more accurate estimate of the positive-order components is obtained by subtracting the temporary update value of the positive-order weights obtained in the previous step and the positive rotation factor in the complex reference vector. This process can be expressed by the formula:

[0033] in, This is the decoupling target signal for the negative-order channel. Within the same computation cycle, this step implements a feedforward flow of information, using the just-optimized positive-order estimation result to feed back into the negative-order estimation process. This ensures that the update of the negative-order weights is based on a better understanding of the system state. Secondly, it calculates the error between the deep decoupling target signal and the current negative-order weights' estimation of the negative-order components, i.e., the negative-order channel decoupling error. Then, according to the LMS algorithm criterion, using the negative-order learning rate... The step size is used to update the current negative order weights. This process can be expressed by the formula:

[0034] in, This is the final updated value of the negative order weights (final updated value of the negative order weights). The learning rate is negative (negative order learning rate). The complex conjugate of the negative-sequence channel decoupling error is used. Thus, by employing feedforward, more precise information to purify the learning target, the convergence process of the negative-sequence weights is greatly accelerated, achieving high-speed and accurate identification of grid imbalance components.

[0035] Furthermore, based on the final updated value of the negative-order weights and the temporary updated value of the positive-order weights, an updated complex weight vector is generated. That is, the temporary updated value of the positive-order weights and the final updated value of the negative-order weights calculated in the previous steps are combined into a complete complex weight vector, which is then used as the initial weight for the next time step. This allows for the output of a weight vector that accurately represents the current state of the positive and negative order components of the power grid, providing a high-fidelity mathematical basis for the subsequent generation of compensation commands.

[0036] This mechanism fundamentally solves the performance bottleneck problem of traditional CLMS methods in dynamic imbalance compensation applications at the algorithm level without increasing the complexity of the system's hardware costs. Specifically, by introducing a sequential error feedforward decoupling mechanism, it completely breaks the transient coupling in the positive and negative order weight update process, making the identification process of the two components approximately independent. This effect further leads to a significant improvement in weight convergence speed, especially under severe operating conditions where the grid imbalance state changes abruptly, enabling faster tracking and locking of the actual phasor of the imbalance component. In addition, this mechanism allows setting independent learning rates for positive and negative order channels, providing an additional optimization dimension for dealing with scenarios under different imbalance degrees. Ultimately, the improved algorithm convergence speed and estimation accuracy enable power quality compensation devices to generate faster and more accurate compensation commands, thereby more effectively suppressing grid voltage imbalance and ensuring stable power quality.

[0037] Specifically, in step S5, a negative-sequence compensation complex command is generated based on the updated complex weight vector. It should be understood that although the preceding adaptive weight update process has accurately identified the complex amplitude of the negative-sequence component in the grid voltage (i.e., the final updated value of the negative-sequence weight), this weight value itself is an internal parameter and cannot be directly used to modulate power switching devices. Therefore, in the technical solution of this application, the information calculated by the algorithm to describe the grid imbalance state (i.e., the updated weight vector) is transformed into a compensation command that can drive power electronic devices to perform actual compensation operations. Specifically, the mathematical representation of the negative-sequence weight is converted into a waveform command that varies in the time domain and has a specific amplitude and phase. After PWM modulation, this waveform command can drive the inverter unit in the UPQC to generate the required compensation voltage or current, ultimately realizing the physical process of imbalance mitigation.

[0038] In specific implementation, firstly, the negative-order component phasor is extracted from the updated complex weight vector to obtain the negative-order component phasor. Here, it should be understood that the core principle of complex linear neural networks (such as ADLINE) applied to power quality analysis is to approximate a desired signal using a linear combination of a weight coefficient vector and an input reference vector (i.e., a complex reference vector). In the technical solution of this application, the complex weight vector corresponds to a complex reference vector containing positive and negative order twitch factors. This means that after a sufficiently large number of iterations, each weight value in the complex weight vector will converge to a value that is exactly equal to the complex amplitude of the component in the desired complex signal (i.e., the complex space vector of the grid voltage) associated with the corresponding reference vector component (positive or negative order twitch factor). Therefore, when the algorithm converges, the steady-state value of the weight associated with the negative order twitch factor in the complex weight vector directly represents the complex amplitude (i.e., phasor) of the negative-order component in the grid voltage. Therefore, based on this theory, this step extracts the weight value representing the negative-order component from the complex weight vector that has converged to a stable state after iterative updates by the adaptive algorithm in each sampling period. This weight value serves as a definite physical quantity that can be used in subsequent steps, namely the negative-order component phasor. This step is the bridge between the internal state of the algorithm and the definite physical quantity, providing the most direct input for generating accurate compensation instructions.

[0039] In this process, the complex reference vector is a column vector with a fixed structure, containing two elements: the first element is a forward twitch factor for matching the positive-sequence component, and the second element is a reverse twitch factor for matching the negative-sequence component. Correspondingly, the updated complex weight vector generated through the update process is also a two-dimensional column vector, containing two elements: the first element is the positive-sequence weight (temporary update value) corresponding to the positive-sequence twitch factor, and the second element is the negative-sequence weight (final update value) corresponding to the negative-sequence twitch factor. According to the Wiener solution in adaptive filtering theory, when the algorithm converges, the optimal value of the weight vector should make its linear combination with the reference vector best approximate the desired signal. This causes the negative-sequence weight to converge to the complex amplitude of the negative-sequence component of the grid voltage. Therefore, the extraction of the negative-sequence component phasor is implemented by directly accessing the second element of the updated complex weight vector, i.e., the final update value of the negative-sequence weight, and assigning it to the variable representing the negative-sequence component phasor. It is worth noting that this process can be accomplished by reading values ​​from specific memory addresses or registers in a digital signal processor (DSP) or field-programmable gate array (FPGA).

[0040] Furthermore, the negative-sequence component phasor is reconstructed in the time domain to obtain the negative-sequence compensation complex command. It should be understood that the basic principle of managing three-phase imbalance is to inject a compensation voltage or current into the power grid through the series or parallel units of the UPQC (Upgraded Quantization Control) unit. This compensation voltage or current has the same amplitude but opposite phase as the detected negative-sequence voltage component, thereby counteracting the imbalance effect. The negative-sequence component phasor extracted in the previous step precisely quantifies the degree (amplitude) and location (phase) of the imbalance, but it is not itself a specific control command. Therefore, in the technical solution of this application, based on the information of the negative-sequence component and combined with the control objective, the complex form of the compensation signal that the UPQC needs to output to achieve this objective is calculated. This complex command is then sent to a PWM (Pulse Width Modulation) generator and ultimately converted into a drive signal controlling the on / off state of power switching devices such as Insulated Gate Bipolar Transistors (IGBTs), thereby physically achieving precise energy compensation.

[0041] In this process, to achieve the goal of canceling the negative-sequence component, the most direct control strategy is to make the compensation amount generated by UPQC equal in magnitude and opposite in direction (phase) to the detected negative-sequence component. Specifically, this can be achieved by inverting the extracted negative-sequence component phasor and calculating its product with the control gain to obtain the negative-sequence compensation complex command. Here, the control gain is an adjustable real-valued parameter, typically set to 1 for complete compensation, but can also be slightly adjusted according to the actual needs of system stability and compensation strength (e.g., slightly less than 1 to maintain a stability margin). The calculation principle of this generation process is: the compensation command equals the negative control gain multiplied by the negative-sequence component phasor. Ultimately, the generated negative-sequence compensation complex command represents the vector representation of the compensation amount required by the UPQC device to cancel the negative-sequence component of the power grid in the complex plane. It fully defines the amount of compensation needed at the current sampling moment to cancel the negative-sequence component. The vector direction and magnitude of the compensation signal generated in the coordinate system.

[0042] In summary, the neural network-based power quality compensation control method according to the embodiments of this application is explained. It transforms the three-phase voltage into a complex spatial vector to retain amplitude and phase information, uses a constructed complex reference vector as a reference, and adaptively iteratively updates the weights based on the deviation between the desired signal and the current state. This method establishes a mechanism that can automatically adjust its learning ability according to changes in grid operating conditions. It can quickly converge to track large signal disturbances during grid voltage dips or sudden changes, while maintaining weight stability to suppress high-frequency noise during steady-state operation. This effectively breaks the constraint between response speed and compensation accuracy, accurately generates negative-sequence compensation commands, and achieves efficient and flexible management of power quality problems in distribution networks.

[0043] Furthermore, a power quality compensation control system based on neural networks is also provided.

[0044] Figure 3 This is a block diagram of a neural network-based power quality compensation control system according to an embodiment of this application. Figure 3 As shown, the neural network-based power quality compensation control system 300 according to an embodiment of this application includes: a voltage value acquisition module 310, used to acquire instantaneous three-phase voltage values; a complex space vector transformation module 320, used to perform complex space vector transformation on the instantaneous three-phase voltage values ​​to obtain a complex voltage space vector, and designate the complex voltage space vector as a desired complex signal; a complex reference construction module 330, used to construct a complex reference vector based on the grid fundamental angular frequency and sampling period; a complex weight update module 340, used to perform adaptive weight update on the complex weight vector based on the complex reference vector and the desired complex signal to obtain an updated complex weight vector; and a negative-order compensation complex instruction generation module 350, used to generate a negative-order compensation complex instruction based on the updated complex weight vector.

[0045] As described above, the neural network-based power quality compensation control system 300 according to the embodiments of this application can be implemented in various wireless terminals, such as servers with neural network-based power quality compensation control algorithms. In one possible implementation, the neural network-based power quality compensation control system 300 according to the embodiments of this application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the neural network-based power quality compensation control system 300 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the neural network-based power quality compensation control system 300 can also be one of many hardware modules of the wireless terminal.

[0046] Alternatively, in another example, the neural network-based power quality compensation control system 300 and the wireless terminal can also be separate devices, and the neural network-based power quality compensation control system 300 can be connected to the wireless terminal via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0047] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A power quality compensation control method based on neural networks, characterized in that, include: Obtain the instantaneous values ​​of the three-phase voltage; The instantaneous values ​​of the three-phase voltages are subjected to complex space vector transformation to obtain complex voltage space vectors, and the complex voltage space vectors are designated as the desired complex signals; A complex reference vector is constructed based on the fundamental angular frequency of the power grid and the sampling period. Based on the complex reference vector and the desired complex signal, the complex weight vector is adaptively updated to obtain the updated complex weight vector; Based on the updated complex weight vector, generate negative order compensation complex instructions.

2. The power quality compensation control method based on neural networks according to claim 1, characterized in that, Performing a complex space vector transformation on the instantaneous values ​​of three-phase voltages to obtain a complex voltage space vector includes: performing a complex space vector transformation on the instantaneous values ​​of three-phase voltages using the following formula: in, These are the instantaneous values ​​of the three-phase voltage. It is a complex voltage space vector.

3. The power quality compensation control method based on neural networks according to claim 1, characterized in that, Based on the fundamental angular frequency of the power grid and the sampling period, a complex reference vector is constructed, including: constructing the complex reference vector using the following formula, where the formula is: in, The sampling period is The fundamental angular frequency of the power grid. It is a positive rotation factor. It is the reverse rotation factor. This is a discrete-time index.

4. The power quality compensation control method based on neural networks according to claim 1, characterized in that, Based on the complex reference vector and the desired complex signal, the complex weight vector is adaptively updated to obtain the updated complex weight vector, including: Based on the complex reference vector and the expected complex signal, the positive order component error prediction and preliminary weight update of the complex weight vector are performed to obtain the temporary updated value of the positive order weight. Based on the expected complex signal and the temporary update value of the positive order weight, the current negative order weight is accurately estimated for the negative order component error and the weight is finally updated to obtain the final update value of the negative order weight. Based on the final update value of the negative-order weights and the temporary update value of the positive-order weights, an updated complex weight vector is generated.

5. The power quality compensation control method based on neural networks according to claim 4, characterized in that, Based on the complex reference vector and the desired complex signal, the positive-sequence component error is predicted and the weights are initially updated to obtain temporary updated values ​​for the positive-sequence weights. This includes: predicting the positive-sequence component error and initially updating the weights of the complex weight vector using the following formula: in, The target signal for decoupling the forward-sequence channel. For the desired complex signal, The current negative weight in the complex weight vector. The reverse rotation factor in the complex reference vector. This is a temporary update value for the orthogonal weights. The current positive order weight in the complex weight vector. For positive learning rate, The positive rotation factor in the complex reference vector. It is the complex conjugate of the decoupling error of the positive sequence channel.

6. The power quality compensation control method based on neural networks according to claim 4, characterized in that, Based on the expected complex signal and the temporary update value of the positive-order weights, the current negative-order weights are precisely estimated for the negative-order component error and the weights are finally updated to obtain the final update value of the negative-order weights. This includes: precisely estimating the negative-order component error and finally updating the weights of the current negative-order weights using the following formula, where the formula is: in, The target signal for decoupling the negative sequence channel. This is the final updated value for the negative order weights. For negative learning rates, It is the complex conjugate of the decoupling error of the negative-order channel.

7. The power quality compensation control method based on neural networks according to claim 1, characterized in that, Based on the updated complex weight vector, negative-order compensation complex instructions are generated, including: The negative-order component phasor is extracted from the updated complex weight vector to obtain the negative-order component phasor. The negative-order component phasor is reconstructed in the time domain by performing a negative-order compensation instruction to obtain the negative-order compensation complex number instruction.

8. A power quality compensation control system based on neural networks, characterized in that, include: Voltage acquisition module, used to acquire instantaneous three-phase voltage values; The complex space vector transformation module is used to perform complex space vector transformation on the instantaneous values ​​of three-phase voltages to obtain complex voltage space vectors, and to designate the complex voltage space vectors as the desired complex signals; The complex reference construction module is used to construct a complex reference vector based on the fundamental angular frequency of the power grid and the sampling period; The complex weight update module is used to adaptively update the complex weight vector based on the complex reference vector and the desired complex signal to obtain the updated complex weight vector. The negative-order compensation complex number instruction generation module is used to generate negative-order compensation complex number instructions based on the updated complex number weight vector.

Images