Harmonics pose a significant hazard to power grids and electrical equipment [1, 2]. Problems caused by harmonic pollution are serious and must be properly addressed. Installing compensation devices to compensate for harmonics and reactive power, and using filters to filter and compensate for distorted currents are important means to solve the problems of harmonics and reactive power interference in the power grid, and are also the ultimate measures to eliminate the impact of harmonic currents on the power grid. The schematic diagram of an active power filter is shown in Figure 1.1. The active filter injects an anti-phase harmonic current ic into the power grid with an amplitude equal to that of the load harmonic current to cancel the original harmonic current, making the power supply current is approximately a sine wave. The result detected by the harmonic detection circuit is the input of the control circuit, which in turn determines the output of the control circuit. The output of the control circuit, in turn, determines the output compensation current ic of the active power filter. Therefore, the magnitude, direction, phase, and accuracy of the compensation current of the active power filter largely depend on the harmonic detection circuit. If the detection circuit has advantages such as good real-time performance, high accuracy, and small error, the compensation of the active power filter will be better. Otherwise, the active power filter compensation current ic is either greater than the harmonic current, resulting in overcompensation and thus becoming another harmonic source; or ic is less than the harmonic current, resulting in undercompensation. Harmonic detection plays a significant role in the compensation of active power filters. If the compensation effect of active power filters is to be good, the harmonic detection circuit must have advantages such as good real-time performance, high accuracy, and small error. Real-time performance and continuity depend on the real-time performance and accuracy of reactive current detection. Therefore, it is necessary to study the harmonic measurement circuit in depth. Traditional detection methods include Fryze time-domain decomposition, slot filter, frequency domain-based FFT transformation method, and synchronous detection method for unbalanced three-phase systems. The main disadvantages of these methods are: 1) large time delay; 2) difficult to implement, such as the slot filter method; 3) large error when the voltage is distorted. The disadvantage of the recently proposed PQ method [3] is that when the power supply distortion is large, the voltage contains distortion, so the calculated iaf, ibf, and icf also contain harmonics. Since they are composed of harmonic current components, the method has a large error when the harmonics are large. 1. Harmonic Detection Method Based on Adaptive Interference Cancellation Principle Adaptive interference cancellation theory is a widely used signal processing technique in recent years [4, 5, 6]. Because it can continuously learn and adjust itself to maintain the system in an optimal state, it has been applied in various fields. The principle of adaptive interference cancellation is as follows: The system has two input terminals—the original input and the reference input (see Figure 2 for the schematic diagram). The original input of the system is the signal S and additive noise N0, which are uncorrelated. The reference input of the system is noise N1, which is correlated with N0 but uncorrelated with S. After N1 is processed by an adaptive filter, it is subtracted from the original input signal. The system is adjusted by an adaptive algorithm to minimize the average output power of the system, that is, to make N1 approximate N0. Then, the N0 component in the original input is subtracted to achieve the purpose of canceling interference. It can be proven that the output ε of the system at this time is the best estimate of the signal S under the minimum mean square criterion [7]. The purpose of harmonic detection is to extract harmonics and reactive current in the distorted current in real time. If the fundamental voltage is used as the reference input, and the nonlinear load current is used as the original input... Similar to the above, the reference input after passing through the adaptive filter is ultimately forced to approximate the fundamental signal in the original voltage input in terms of amplitude and phase. Then, this fundamental component is subtracted from the load current, resulting in the system output—the sum of all harmonic components and reactive power—thus achieving the purpose of detecting harmonics and reactive power. Similarly, using the fundamental wave and its orthogonal function as input achieves the purpose of detecting harmonic current. 2. Harmonic Detection Method Based on Artificial Neuron Adaptation 2.1 Time-Delay-Free Artificial Neuron Harmonic and Reactive Power Detection Method A single artificial neuron has certain processing, calculation, and mapping capabilities, and possesses adaptive and self-learning abilities. Therefore, the detection system implemented by the adaptive filter mentioned above can also be replaced by neurons. If the performance is good and the learning algorithm is simple, the simple neuron structure can achieve the purpose of rapid harmonic detection. The time-delay-free harmonic detection method can detect the generalized reactive current ib (that is, the sum of harmonic current ih and reactive current i1q). The block diagram of the time-delay-free adaptive neuron harmonic and reactive current detection method is shown in Figure 3. 2.2 Neuron Adaptive Harmonic Detection Method with Orthogonal Function Input This method is actually a special case of the neuron harmonic detection method with time delay. That is, the time delay is set to -T/4, m is set to 1, and two orthogonal functions are formed as reference inputs. In implementation, the time delay of this method is set to -T/4. Its implementation can take the voltage sample of a quarter cycle before the current time stored in memory as the cosine function. If the voltage frequency is relatively constant (generally the voltage distortion is very small), the difference in the number of samples can be calculated first, and then the current sampled voltage and the sampled voltage calculated before the current time stored in memory can be used as reference input. Then, the reference input vector is: X(k)=[u(k),u(k+T/4)] (1) When the voltage frequency does not change much, the time before the current time can be calculated based on the sampling frequency and the time required for the fundamental wave to shift by 900 phase angle: T/4¸T′=1/(50´4)¸(1/2000)=10 (2) Where T is the period of the voltage with a fixed frequency, and T′ is the sampling period. Figure 4 is a block diagram of the neuron adaptive harmonic detection method with a fixed voltage frequency and orthogonal function input. 2.3 Implementation of Harmonic Detection Method Based on Artificial Neuron Adaptation Here, the neuron is a discrete input, the weight w is an analog quantity, and a random value is taken between [-1, 1]. The sampling period is 2000 Hz, the initial threshold is zero, the learning threshold is initially selected as h = 0.11, and then adjusted according to the experimental results. The inertia coefficient is a = 0.1. The arrays x (120) and y (120) are the voltage and current sampling values ​​stored in the PC memory, respectively. 3 Experimental Results This experiment uses a Pentium CPU, a PC compatible machine with a main frequency of 100 MHz and 8 MB of memory as the hardware for data calculation, and the PCL-818L multifunction card as the hardware for data sampling, A/D conversion, and data transmission. Since the unit square wave is representative, this paper uses the unit square wave nonlinear load current for the experiment. 3.1 Experimental Results of Time-Delay-Free Artificial Neuron Harmonic and Reactive Current Detection Method 3.1.1 Experimental Results of Single-Square-Wave Nonlinear Load Current Through extensive experimental research, it was found that the calculated results differed from the experimental results for different h values. When h = 0.04 and a = 0.01, ir approached i1p, id approached ic, and the error was very small, as shown in Figure 5. Figure 5 shows the initial weights as random values ​​between -1 and 1, with an initial threshold of zero. Using the power supply voltage as a reference input and a sampling frequency of 2000Hz (2000Hz for all experiments below), the experimental results of calculating the harmonic and reactive current of a square-wave nonlinear load current with an amplitude of 1 were obtained. To understand and verify the experimental results, the waveforms of ic (detected value) and i1p (theoretically calculated) were plotted simultaneously in the figure. Figure 5.b shows the active current separated from Figure 5.a, and the waveforms of harmonic and reactive current are shown in Figure 5.c. Figure 5.d shows the error between the detected harmonic and reactive current and the theoretical value. Figure 5 clearly shows that in the first cycle, both the active current, harmonic current, and reactive current deviate from the theoretical values. Around the second cycle, these differences become less significant. The calculated active current ir approaches its theoretical value i1p, while the calculated harmonic and reactive current id approaches the theoretical generalized reactive current ic. Their errors after the second cycle are only below 0.1A, accounting for about 5% of the total current. These results indicate that a reasonable selection of the learning rate h of the neuron has a significant impact on the learning convergence of the neuron; a reasonable selection of the learning rate h can accelerate convergence. 3.1.2 Experimental results of the frequency-varying square wave nonlinear load current are detailed in Figure 6. These results are obtained under the conditions of h = 0.04 and a = 0.01, when the system is restarted and the frequency of the load current is changed from 50Hz to 55Hz in the third cycle. Figures 6.b, 6.c, and 6.d clearly show that frequency has little impact on the no-delay neuron harmonic detection system. This means the method is highly adaptable to frequency changes, almost completely following them. Active current, harmonics, and reactive current all change with frequency, and the error rate remains unchanged. 3.2 Neuron Adaptive Harmonic Detection Method with Orthogonal Function Input 3.2.2 Square Wave Detection Results for Load Current Variation See Figure 7 for details. Figures 7.b and 7.c show that when the load current abruptly changes from 1A to 0.5A, after approximately one cycle of adaptation, the calculated fundamental current ir and harmonic current id approach their theoretical values ​​i1 and ih, indicating that this method also has good tracking ability for load current variations. From the error curve in Figure 7.d, the error is less than 0.1A in the second cycle of the load current variation. 3.2.3 Detection Results of Square Wave Nonlinear Load Current with Frequency Variation As can be clearly seen from Figure 8, the neuron adaptive harmonic detection system using orthogonal functions as reference input is not "sensitive" to frequency changes. It does not produce abrupt changes when the frequency changes and can quickly follow the frequency variations. In Figures 8.b and 8.c, the calculated fundamental current ir and harmonic current id closely change with their theoretical currents i1 and ic. Therefore, the system always has the ability to automatically follow the frequency of the detected load current. 4 Conclusion In summary, the experimental results prove the correctness of the proposed time-delay-free neuron adaptive harmonic and reactive current detection method, as well as the orthogonal input neuron adaptive harmonic current detection method. The proposed method is adaptive to the frequency and amplitude changes of the load current, proving that the proposed method has good real-time performance. Therefore, the method proposed in this paper is indeed feasible. References: [1] Joseph S.Subjak et al., Harmonics-causes, effects, measurements, and analysis; An Update, IEEE Trans. on Ind. April, VOL.26, NO.6, 1034-1042, Nov./Dec. 1990 [2] DVBose, Harmonics analysis and suppression for electrical system supplying static converter and other nonlinear loads, IEEE Trans. on Ind. April, vol.15, NO.5, 1979 [3] Tang Hongcheng, Zhang Xiaoqing, Feng Puqiao, Current status and prospect of real-time detection methods for harmonics and reactive current, Journal of Logistics Engineering College, VOL.14, NO.3, 48-53, 1998 [4] Luo Shiguo, Research on active power filters, Doctoral dissertation, Chongqing University, 1-2, 1993 [5] John R. Glover, Adaptive Noise Canceling Applied to Sinusoidal Interference, IEEE Trans. on Accost, Speech, and Signal Processing, Vol.ASSP-25, 484-491, Dec. 1977 [6] CFN Cowan and PM Grant, Adaptive Filters, Prentice-Hall, Inc., 1985 [7] Tang Hongcheng, Research and Implementation of Digital Detection Method for Harmonic Current, Master's Thesis, Logistics Engineering College, 18-20, 1999 Editor: He Shiping