Method for intelligently adjusting generator based on AI

Abstract

The invention relates to the technical field of automatic adjustment of synchronous generators, in particular to a method for intelligently adjusting a generator based on AI, which comprises the following steps of: obtaining positive half-cycle maximum and negative half-cycle minimum excitation current time judgment offset, obtaining a transient offset conclusion, extracting a variation amplitude by adopting a voltage sequence, evaluating a voltage condition by combining the offset conclusion, and calculating the voltage condition. And comparing the frequency power change rate to generate an adjusting signal, and updating the recursive excitation reference value sequence. According to the method, the extreme value time points of the excitation current in the positive and negative half cycles are obtained, the offset is judged in combination with the cycle length, the abnormity is quickly revealed, the voltage sequence is collected, the adjacent sampling difference amplitude is extracted and corresponds to the offset result, cross comparison is carried out in combination with the frequency and the power rate, and multi-parameter collaborative judgment adjustment triggering is driven. A dynamic instruction sequence is generated according to the excitation current rate difference value and the trigger state, synchronous adjustment of transient unbalance and the period trend is achieved, the rapid response and continuous stability capacity is enhanced, and the problems of over-adjustment and frequent start and stop are solved.

Description

Technical Field

[0001] This invention relates to the field of automatic synchronous generator technology, and in particular to a method for intelligent generator regulation based on AI. Background Technology

[0002] The field of automatic regulating synchronous generator technology involves technical solutions for real-time monitoring and regulation of synchronous generator operating parameters to maintain stable operation. This field mainly includes the detection and regulation of electrical characteristics such as voltage, frequency, and speed of the synchronous generator, as well as its mechanical state. By controlling the excitation current or load-side regulation, it ensures that the generator outputs power quality that meets the requirements of the power grid or load under different operating conditions. The overall method is usually based on real-time acquisition of generator operating characteristics and dynamic adjustment of control quantities in combination with regulation strategies to achieve closed-loop control. Among them, the traditional automatic regulating synchronous generator refers to using an automatic voltage regulator to adjust the excitation current by preset voltage or frequency reference values ​​to maintain stable operation of the synchronous generator under rated voltage and frequency conditions. The method used is usually based on PID control algorithm and steady-state error correction link. Control is achieved by detecting the generator output voltage deviation and adjusting the excitation device. This method mainly relies on voltage and current transformers to collect generator output side signals and excitation current controllers to adjust the excitation winding current to achieve the control purpose.

[0003] Existing technologies for automatically adjusting synchronous generators mainly rely on fixed preset voltage or frequency reference values. They drive excitation regulation by detecting steady-state deviations. While this can maintain basic stability under rated operating conditions, it relies too heavily on steady-state deviation control and lacks real-time analysis of transient excitation current deviations and the coupling of multiple parameters such as voltage, frequency, and power. This makes it prone to response lag under rapid load fluctuations or nonlinear disturbances, resulting in amplified voltage and frequency fluctuations that are difficult to return to a stable range in a short time. This leads to phenomena such as grid voltage flicker and excessive reactive power surges. In practice, it is common for the excitation device to detect deviations under sudden load increases or decreases, but because the control algorithm only corrects steady-state errors, it fails to identify the rapid coupling relationship between current deviation and voltage and frequency rates in a short time. This results in delayed or unbalanced adjustment commands, causing the generator output to deviate further from the target operating range, increasing the amplitude and frequency of subsequent adjustments, and also increasing the risk of wear and tear on mechanical and electrical components. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an AI-based intelligent generator regulation method.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for intelligently regulating a generator based on AI, comprising the following steps:

[0006] S1: Obtain the maximum sampling time point of the excitation current in the positive half-cycle and the minimum sampling time point in the negative half-cycle within the cycle. Use the relationship between these two time points and the cycle length to determine the degree of deviation and obtain the transient excitation deviation judgment conclusion.

[0007] S2: Based on the transient excitation offset judgment conclusion, the output voltage sequence of the corresponding period is collected, the change amplitude between adjacent sampled values ​​is extracted, and then the condition judgment is performed through the correspondence with the excitation offset conclusion to form the voltage change condition evaluation result;

[0008] S3: Based on the voltage change condition evaluation results, collect the frequency fluctuation sequence and power factor sequence of the same period, obtain the change rate of adjacent sampling points, and compare the combination relationship between frequency and power change rate with the voltage change condition to obtain the intelligent adjustment trigger signal;

[0009] S4: Call the intelligent adjustment trigger signal to obtain the maximum rate value of the full sequence of periodic excitation current in the positive and negative half-cycles and perform direct difference calculation. Compare the obtained rate difference with the trigger state of the intelligent adjustment trigger signal to generate a dynamic command sequence for excitation adjustment.

[0010] As a further aspect of the present invention, the transient excitation offset judgment conclusion includes offset time difference, period ratio coefficient, and offset degree level; the voltage change condition evaluation result includes voltage change amplitude threshold, voltage change trend type, and voltage condition compliance status; the intelligent adjustment trigger signal includes adjustment trigger flag, adjustment execution priority, and trigger validity status; and the excitation adjustment dynamic command sequence includes rate difference reference, command action sequence number, and command applicable period segment.

[0011] As a further aspect of the present invention, the specific steps of S1 are as follows:

[0012] S101: Obtain the sampling sequence of the excitation current during the positive and negative half-cycles within the cycle. Based on the relationship between the current value and time point of each sampling point in the positive half-cycle, filter the corresponding time point with the largest current value to obtain the peak time point of the positive half-cycle.

[0013] S102: Call the peak time point of the positive half-cycle, and based on the relationship between the current value and the time point of the sampling point of the negative half-cycle, filter the corresponding time point with the smallest current value, and combine the time point with the interval between the peak time point of the positive half-cycle to obtain the excitation time difference.

[0014] S103: Based on the ratio of the excitation time difference to the cycle length, and according to the preset offset judgment benchmark, determine the offset degree range and obtain the transient excitation offset judgment conclusion.

[0015] As a further aspect of the present invention, the specific steps of S2 are as follows:

[0016] S201: Obtain the transient excitation offset judgment conclusion, collect the corresponding period output voltage sequence, extract the change amplitude of adjacent sample values ​​in the sequence in turn, and generate adjacent sample change sequence;

[0017] S202: Based on the adjacent sampling change sequence, call the offset amplitude reference value in the excitation offset judgment conclusion, complete the comparison between the change amplitude and the reference value, obtain the proportion exceeding the reference value, and obtain the offset amplitude proportion value;

[0018] S203: Based on the offset amplitude ratio and the judgment threshold in the excitation offset judgment conclusion, a condition judgment is made to form a normal or abnormal classification result, and the voltage change condition evaluation result is obtained.

[0019] As a further aspect of the present invention, the specific steps of S3 are as follows:

[0020] S301: Based on the voltage change condition evaluation results, obtain the frequency fluctuation sequence and power factor sequence of the same period, call the sampling point data of the voltage change condition, deduce the change rate of adjacent sampling points, and generate frequency and power factor change rate sequences.

[0021] S302: Based on the frequency and power factor change rate sequence, call the voltage change data of the voltage change condition sequence, compare the frequency change rate, power factor change rate and voltage change value in sequence order, filter the points that simultaneously meet the threshold conditions, and obtain the threshold correlation sequence.

[0022] S303: Based on the threshold association sequence, identify the range of points that continuously meet the conditions, determine the starting position of the range as the trigger reference, and obtain the intelligent adjustment trigger timing signal.

[0023] As a further aspect of the present invention, the rate of change of the sampling interval refers to obtaining the degree of change of the sequence over time by dividing the difference in sequence values ​​between two adjacent sampling points by the sampling time interval, given the time interval data of each sampling point.

[0024] The power factor change rate refers to the time-varying behavior of the power factor sequence, specifically within the derived change rate.

[0025] As a further aspect of the present invention, the specific steps of S4 are as follows:

[0026] S401: Obtain the complete sequence of the periodic excitation current of the intelligent adjustment trigger signal, divide it into positive half-cycle current sequence and negative half-cycle current sequence according to the period boundary point, extract the change rate information of the two sequences, and generate positive and negative half-cycle rate sequences.

[0027] S402: Based on the positive and negative half-cycle rate sequences, detect the maximum value in the positive half-cycle rate and the maximum value in the negative half-cycle rate, calculate the difference between the two, and obtain the maximum rate difference.

[0028] S403: Based on the maximum rate difference and the triggering state of the intelligent adjustment trigger signal, the condition is determined and filled into the corresponding position of the dynamic sequence. After completing the process, the excitation adjustment dynamic command sequence is obtained.

[0029] As a further aspect of the present invention, the positive half-cycle current sequence is a sequence formed by selecting continuous data segments in the positive direction within the cycle from the obtained full sequence of periodic excitation current based on the cycle boundary point.

[0030] The negative half-cycle current sequence is also derived from the full cycle excitation current sequence. The continuous data segments with negative current are extracted through the cycle boundary points to form an independent sequence.

[0031] As a further aspect of the present invention, the method further includes:

[0032] S5: Based on the excitation adjustment dynamic command sequence, update the adjustment reference value according to the command sequence in combination with the excitation current fluctuation characteristics to generate a periodic recursive excitation reference value sequence.

[0033] The periodic recursive excitation reference value sequence includes recursive update values, periodic markers, and excitation target values.

[0034] As a further aspect of the present invention, the specific steps of S5 are as follows:

[0035] S501: Based on the excitation regulation dynamic command sequence, extract the target regulation reference value and the corresponding time point, and compare the numerical relationship of adjacent reference values ​​in the time point sequence in turn to obtain the reference value change range sequence;

[0036] S502: Call the reference value change range sequence and the instantaneous value of the excitation current at the corresponding time point, compare the reference value change range with the excitation current fluctuation range at the time point, select the time point that exceeds the excitation current fluctuation range threshold, and obtain the over-amplitude time index set.

[0037] S503: Locate the corresponding time point of the original target adjustment reference value sequence according to the overamplitude time index set, perform incremental adjustment on the corresponding reference values ​​in sequence, and combine them into a sequence according to the time order to obtain the periodic recursive excitation reference value sequence.

[0038] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0039] In this invention, by acquiring the extreme time points of the excitation current in the positive and negative half-cycles and judging the offset in combination with the cycle length, anomalies are quickly revealed. The voltage sequence is collected, the difference between adjacent samples is extracted and matched with the offset results, and voltage fluctuations are sorted out. The frequency and power rate are cross-compared to drive multi-parameter collaborative judgment and adjustment triggering. A dynamic command sequence is generated based on the difference in excitation current rate and triggering state, and the reference value is recursively updated to achieve synchronous adjustment of transient imbalance and cycle trend, enhance the ability of rapid response and continuous stability, and alleviate the problems of over-adjustment and frequent start-stop. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the steps of the present invention;

[0041] Figure 2 This is a flowchart of steps S1 of the present invention;

[0042] Figure 3 This is a flowchart of steps S2 of the present invention;

[0043] Figure 4 This is a flowchart of steps S3 of the present invention;

[0044] Figure 5 This is a flowchart of step S4 of the present invention;

[0045] Figure 6 This is a flowchart of step S5 of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0047] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0048] Please see Figure 1 A method for intelligently regulating a generator based on AI includes the following steps:

[0049] S1: Obtain the maximum sampling time point of the excitation current in the positive half-cycle and the minimum sampling time point in the negative half-cycle within the cycle. Use the relationship between these two time points and the cycle length to determine the degree of deviation and obtain the transient excitation deviation judgment conclusion.

[0050] S2: Based on the transient excitation offset judgment conclusion, the output voltage sequence of the corresponding period is collected, the change amplitude between adjacent sampled values ​​is extracted, and then the condition judgment is performed through the correspondence with the excitation offset conclusion to form the voltage change condition evaluation result;

[0051] S3: Based on the voltage change condition evaluation results, collect the frequency fluctuation sequence and power factor sequence of the same period, obtain the change rate of adjacent sampling points, and compare the combination relationship between frequency and power change rate with the voltage change condition to obtain the intelligent adjustment trigger signal;

[0052] S4: Call the intelligent adjustment trigger signal, obtain the maximum rate value of the full sequence of periodic excitation current in the positive and negative half-cycles, perform direct difference calculation, compare the obtained rate difference with the trigger state of the intelligent adjustment trigger signal, and generate the excitation adjustment dynamic command sequence.

[0053] S5: Based on the dynamic command sequence of excitation regulation, the regulation reference value is updated according to the command sequence in combination with the characteristics of excitation current fluctuation, and a periodic recursive excitation reference value sequence is generated.

[0054] The transient excitation offset judgment conclusions include offset time difference, period proportional coefficient, and offset degree level; the voltage change condition evaluation results include voltage change amplitude threshold, voltage change trend type, and voltage condition compliance status; the intelligent adjustment trigger signal includes adjustment trigger flag, adjustment execution priority, and trigger validity status; the excitation adjustment dynamic command sequence includes rate difference reference, command action sequence number, and command applicable period segment; and the periodic recursive excitation reference value sequence includes recursive update value, period marker, and excitation target value.

[0055] Please see Figure 2 The specific steps of S1 are as follows:

[0056] S101: Obtain the sampling sequence of the excitation current during the positive and negative half-cycles within the cycle. Based on the relationship between the current value and time point of each sampling point in the positive half-cycle, filter the corresponding time point with the largest current value to obtain the peak time point of the positive half-cycle.

[0057] When acquiring the sampling sequence of excitation current within a cycle, it is typically done using a current sensor configured in the motor driver, such as a Hall sensor or shunt resistor, combined with a high-frequency ADC for current detection. With a sampling frequency of 10kHz and a cycle length of 20ms, approximately 200 current values ​​and corresponding time values ​​can be obtained per cycle. First, all sampled data is divided into positive and negative half-cycles, determined by current values ​​greater than 0 for positive half-cycles and less than 0 for negative half-cycles. For example, starting from 0ms, current values ​​are continuously collected as positive until 9.8ms, and from 9.8ms to 20ms, current values ​​are all negative, forming positive and negative half-cycle sequences. Then, within the positive half-cycle, the current value at each sampling point is compared sequentially. If a detected value is greater than the currently recorded maximum value, the record for the current maximum value is replaced, and the corresponding time is updated synchronously. For example, with a load of 5 Nm, the current reaches 4.2 A at 3.6 ms, which is the maximum current value and its corresponding time for this positive half-cycle. Further, when the load increases to 8 Nm, the maximum current value of 4.7 A is obtained by the same method at 3.8 ms. This process can be continuously executed to accumulate sufficient data for the motor's operation under different load conditions. Finally, it is confirmed that the peak value of the positive half-cycle occurs at 3.6 ms in this cycle.

[0058] S102: Call the peak time point of the positive half-cycle, and based on the relationship between the current value and the time point of the sampling point of the negative half-cycle, filter the corresponding time point with the smallest current value, and combine the time point with the interval between the peak time point of the positive half-cycle to obtain the excitation time difference.

[0059] After determining that the peak value of the positive half-cycle occurs at 3.6ms, the process then shifts to finding the position of minimum current value in the current sequence of the negative half-cycle. By scanning point by point, the current value at each sampling point is compared. If the current value is less than the recorded minimum, the record and its time are updated. For example, if the negative half-cycle starts at 9.8ms, and the current reaches -4.5A at 13.8ms, which is smaller than all previous detected values, this time is recorded as the point of minimum current occurrence in the negative half-cycle. The time difference from the peak value of the positive half-cycle to this point is then calculated. Subtracting 3.6ms from 13.8ms yields 10.2ms, representing the interval between the positive and negative peaks. For a period of 20ms, the ratio of the interval to the period is 51%. If the load is 7Nm, a negative peak might be detected at 13.5ms with a current of -4.8A, resulting in a time difference of 9.9ms and a ratio of approximately 50%. This process can be executed multiple times under different load conditions. By recording the changes in the positive and negative peak time difference under different loads, more operating data can be accumulated, and finally the excitation time difference value of 10.2ms is output for this cycle.

[0060] S103: Based on the ratio of excitation time difference to cycle length, and according to the preset offset judgment benchmark, determine the offset degree range and obtain the transient excitation offset judgment conclusion.

[0061] After obtaining a cycle time difference to cycle length ratio of 51%, and referring to the offset judgment benchmark established through multiple no-load and standard load tests, the range of 0 to 30% is defined as slight offset, 30% to 60% as moderate offset, and above 60% as severe offset. Therefore, 51% here falls into the moderate offset range. The judgment method is that if the ratio is between 0 and 30%, it is directly marked as slight offset; if it is between 30% and 60%, it is moderate offset; and above 60% is severe offset. This offset benchmark is usually adjusted according to the standard value given by the motor manufacturer. Alternatively, it can be determined by the offset ratio statistically analyzed under no-load operation, which is concentrated in the range of 20% to 30%, thus setting the upper limit of slight offset at 30% and the upper limit of moderate offset at 60%. In this scenario, the cycle offset level is judged as moderate offset. If a subsequent detection shows a cycle time difference of 14.4ms, with a ratio of 72%, it directly falls into the severe offset range; if the detection shows 5ms, with a ratio of 25%, it is classified as slight offset. By continuously monitoring the excitation offset under different operating conditions, the offset level can be quickly determined, and this cycle was therefore identified as a moderate offset.

[0062] Please see Figure 3 The specific steps of S2 are as follows:

[0063] S201: Obtain the transient excitation offset judgment conclusion, collect the corresponding period output voltage sequence, extract the change amplitude of adjacent sample values ​​in the sequence in turn, and generate adjacent sample change sequence;

[0064] To determine transient excitation offset in voltage signals during transformer operation, it is necessary to first continuously collect the output voltage cycle sequence based on the power grid online monitoring platform. Typically, sampling is performed every 20 milliseconds, continuously acquiring 200 sampling points to form a complete cycle sequence, with values ​​such as 220.1, 220.4, 220.8, 220.9, and 219.7. After sampling, the variation amplitude between adjacent sample values ​​is extracted. The variation amplitude sequence is obtained by subtracting the previous value from the next. For example, from 220.1 to 220.4, it is calculated as 0.3V; from 220.4 to 220.8, it is 0.4V; and then to 220.9, it is 0.1V. This process is continued, ultimately yielding 199 variation amplitude values. To verify the statistical significance of these amplitudes, the average amplitude is obtained by summing the values ​​point by point and dividing by the number of values. The statistical fluctuation range is calculated by taking the square root of the arithmetic mean of all the differences from the average value. This is expressed numerically as an average of approximately 0.15V and a fluctuation range of approximately 0.12V. If there are individual amplitudes such as 0.6V, these amplitudes far exceed the range of the vast majority of data points. In practice, monitoring conditions are set in the voltage monitoring software. When the amplitude exceeds the warning value (e.g., 0.5V), the corresponding time series index and timestamp are immediately recorded, and these records are saved to the power grid monitoring database for historical traceability. This process can utilize a general data acquisition card and accompanying software to import data files in batches. The data management module compares the average value with the fluctuation range to define the normal range. For example, the fluctuation range is defined as between -0.21V and 0.51V, calculated as the average value plus or minus three times. If the amplitude falls outside this range, an anomaly is marked in the monitoring record. The complete process covers data access, value-by-value processing, over-limit detection, and historical record generation.

[0065] S202: Based on the adjacent sampling change sequence, call the offset amplitude benchmark value in the excitation offset judgment conclusion, complete the comparison between the change amplitude and the benchmark value, obtain the proportion of the deviation exceeding the benchmark value, and obtain the offset amplitude proportion value.

[0066] The specific calculation formula for comparing the change range with the benchmark value is as follows:

[0067]

[0068] Calculate the offset composite index, compare the change range with the benchmark value, obtain the proportion exceeding the benchmark value, and get the offset range proportion value.

[0069] Among them, M Δ S represents the composite degree of the change magnitude and the deviation from the benchmark value. i B represents the amplitude value of the i-th sampling point in the adjacent sampling change sequence. i This represents the i-th offset amplitude reference value in the excitation offset judgment conclusion. This represents the average amplitude value of all sampling points in an adjacent sampling change sequence. The value represents the average of all benchmark values ​​in the excitation offset judgment conclusion, and n represents the total number of sampling points. This represents the summation operation over all sampling points from i=1 to i=n, ​​and |·| represents the absolute value operation.

[0070] S i The amplitude value of the i-th sampling point, in millivolts (mV), is acquired in real time from the target device through a high-precision data acquisition system at a sampling rate of 1kHz.

[0071] B i The reference amplitude value corresponding to the i-th sampling point is expressed in millivolts (mV). The average value of each sampling point is calculated by statistically analyzing the historical data of the equipment under normal operating conditions.

[0072] Sampling sequence S i The average value, in millivolts (mV), is obtained by applying data to all S... i The result is obtained by summing the results and then dividing by the total number of sampling points, n.

[0073] Reference sequence B i The average value, in millivolts (mV), is obtained by analyzing all B values. i The result is obtained by summing the results and then dividing by the total number of sampling points, n.

[0074] n: Total number of sampling points, set to 5 in this example, with a sampling interval of 1 millisecond;

[0075] Actual collected data:

[0076] Sampling point numbers (i): 1, 2, 3, 4, 5;

[0077] S i (mV): 1.2, 1.5, 1.3, 1.6, 1.4;

[0078] B i (mV): 1.0, 1.2, 1.1, 1.3, 1.2;

[0079] Calculation process:

[0080] calculate and

[0081]

[0082] Each term in the formula is calculated individually:

[0083] For i = 1:

[0084] S1-B1=1.2-1.0=0.2mV;

[0085]

[0086] For i = 2:

[0087] S2-B2=1.5-1.2=0.3mV;

[0088]

[0089]

[0090] For i = 3:

[0091] S3-B3=1.3-1.1=0.2mV;

[0092]

[0093] For i = 4:

[0094] S4-B4=1.6-1.3=0.3mV;

[0095]

[0096] For i = 5:

[0097] S5-B5=1.4-1.2=0.2mV;

[0098]

[0099] Sum all terms and calculate the average:

[0100] Results Explanation:

[0101] The result indicates that the average offset complexity of the current sampled sequence relative to the reference sequence is approximately 0.2993 mV. This value reflects the overall offset between the sampled signal and the reference signal; a larger value indicates a more significant offset. This result can be used to further analyze the system's stability or detect anomalies.

[0102] The calculation logic of this formula lies in quantifying the offset characteristics between adjacent sampling amplitudes and the reference amplitude through two parts, and then linearly superimposing them to construct a comprehensive offset index. Firstly, the difference term S is used... i -B i It directly describes the instantaneous offset between the amplitude of the current sampling point and the reference value, and then uses the product term. This reflects the two-way coupling relationship between the sampled value and its overall mean, as well as the benchmark value and its overall mean. This coupling term reflects the consistency or divergence of relative trends, while dividing by |S i -B i The operation of | normalizes the coupling value to the single-point offset scale to avoid numerical imbalance. Then, the difference term is added to the normalized coupling value to form a composite feature value that includes both the single-point absolute deviation and the overall trend coupling. The absolute value operation ensures positive vectorization to avoid the cancellation of positive and negative values. Finally, all sampling points are summed and averaged to calculate the average offset composite index that represents the overall offset degree of the sequence. Thus, a dynamic adjustment effect of the global mean is introduced on the basis of a single instantaneous difference, so that the calculation reflects the transient offset while taking into account the overall fluctuation correlation.

[0103] The offset composite index represents the average deviation of the current sampled sequence from the reference sequence. This value not only directly measures the single-point amplitude difference of each sample point relative to the reference point, but also combines the offset coupling relationship between the sampled sequence and the reference sequence relative to their own mean. Thus, it reflects the comprehensive performance of the consistency or dispersion of local instantaneous fluctuations and global fluctuations. The larger the value of this index, the more significant the overall change of the sampled signal relative to the reference state. It includes both the intensity of single-point offset and the degree of matching or divergence of fluctuation trends, and is an important numerical basis for describing the degree of offset and dynamic characteristics of the system.

[0104] S203: Based on the offset amplitude ratio and the judgment threshold in the excitation offset judgment conclusion, condition judgment is performed to form a normal or abnormal classification result and obtain the voltage change condition evaluation result.

[0105] After obtaining the excess ratio, it needs to be compared with the judgment threshold configured in the excitation offset judgment conclusion to generate the final classification result. This judgment threshold is generally derived from transformer management or industry operation and maintenance experience. For example, based on more than 100 detection data in the past two years, the average excess ratio is about 9%, and with the fluctuation value, the threshold can be reasonably set at 10%. During the comparison execution, the operation and maintenance management system will directly read the judgment threshold configuration. If the current excess ratio is detected to be 11%, it is judged to exceed the threshold, and the word "abnormal" will be output in the status bar; otherwise, if the ratio is 8%, "normal" will be output. In the data processing engine, this judgment logic is often completed through simple conditional judgment blocks, such as based on Python data pipelines or through relational database triggers to execute classification and write form fields. Before setting the threshold, typical samples can be screened with the help of historical operation databases, and the offset rate of different time periods can be manually calculated to form a reasonable range and determine the final value. For example, with an average value of 9% and a fluctuation range of 1.5%, the average plus the fluctuation is selected to be about 10% as the long-term operation and maintenance judgment standard. After the results are generated, the corresponding periodic detection records are marked as normal or abnormal and written back to the monitoring database for subsequent large-scale screening.

[0106] Please see Figure 4 The specific steps of S3 are as follows:

[0107] S301: Obtain the frequency fluctuation sequence and power factor sequence of the same period, call the sampling point data of voltage change conditions, derive the change rate of adjacent sampling points, and generate frequency and power factor change rate sequences.

[0108] When acquiring the frequency fluctuation sequence and power factor sequence for the same period, data acquisition terminals or distribution automation devices are typically configured at power plants or substations. By continuously collecting data for 10 minutes with a pre-set sampling period of 1 second, 600 sets of frequency and power factor values ​​can be obtained. These values ​​are stored sequentially according to their sequence numbers. Then, by reading the voltage value sequence for the same time period and precisely matching it by timestamp, the three sets of data are assembled into a single spreadsheet or database record. This provides the data foundation for further deducing the rate of change of adjacent sampling points. During the deduction process, the frequency value of the current sequence number is read sequentially, and the difference is divided by the sampling period to obtain the frequency change rate. The same steps are performed on the power factor, using Python... Scripts like `n` can generate new rate sequence lists in batches. For example, if the frequency change between point 5 and point 6 is 0.02Hz and the sampling period is 1s, the rate of change is directly 0.02Hz / s. If the power factor changes from 0.85 to 0.87, the rate of change is 0.02 / s. Using this method, all data points can be traversed in a very short time to form a set of 599 frequency change rate and power factor change rate sequences. In practice, NumPy or Pandas are often used to automatically perform calculations on all rows at once. After generating the rate sequences, two columns can be added to the table to store these values, while retaining the correspondence with the original sampling sequence number. Subsequent steps will be based on these complete rate sequences for further filtering.

[0109] S302: Based on the frequency and power factor change rate sequence, call the voltage change data of the voltage change condition sequence, compare the frequency change rate, power factor change rate and voltage change value in sequence, filter the points that simultaneously meet the threshold conditions, and obtain the threshold correlation sequence.

[0110] After generating the frequency and power factor change rate sequences, the voltage change values ​​sampled in that segment need to be retrieved. This is done by subtracting adjacent values ​​from the original voltage sequence to obtain the voltage change sequence, forming a triplet data structure. Then, the rate and voltage change values ​​of each sampling point are compared row by row according to the sequence number. Typically, threshold values ​​are first set: the frequency change rate threshold is usually 0.015 Hz / s, the power factor change rate threshold is 0.015 / s, and the voltage change threshold is 0.3V. These thresholds can be calculated based on ratios of approximately 0.03% and 1.5% for a standard power grid range of 50 Hz and 0.9 power factor, respectively. The voltage is converted to 0.003% for a 10kV level. Subsequently, each sampling point is checked row by row. If the frequency change rate is greater than 0.015 / s... If the frequency change rate is greater than 0.015 Hz / s, the power factor change rate is greater than 0.015 / s, and the voltage change is greater than 0.3V, then the point meets the screening requirements. For example, if the frequency change rate is 0.017 Hz / s, the power factor change rate is 0.02 / s, and the voltage change is 0.4V in the 20th sampling point, then the point number is recorded and added to the threshold association sequence. This process continues until all comparisons are completed. In this way, a sequence list containing only the point numbers that meet the conditions can be obtained, such as numbers 20, 35, 70, etc. In actual implementation, pandas Boolean indexes or database SQL statements are usually used for batch filtering, which can quickly lock all point numbers that exceed the set threshold and form the threshold association sequence required for subsequent analysis.

[0111] S303: Based on the threshold correlation sequence, identify the range of points that continuously meet the conditions, determine the starting position of the range as the trigger reference, and obtain the intelligent adjustment trigger timing signal;

[0112] When analyzing the obtained threshold-related sequence, it is necessary to identify the sampling segments that continuously meet the conditions. By iteratively judging whether the difference between each point number and its next number is 1, if the number of consecutive points exceeds the set number, it is determined to be a valid continuous interval. For example, if the continuous point threshold is set to 5, when the consecutive numbers 50, 51, 52, 53, 54, and 55 are detected in the sequence, that is, the length reaches 6, then 50 is identified as the starting point of the segment. If the scan continues and finds that 80, 81, 82, and 83, although there are only 4 points, are below the set threshold of 5, the segment is not recorded. In this way, occasional or short-term changes can be filtered out, and only continuous and stable trigger reference points are retained. Usually, the threshold length can be selected according to the characteristics of the equipment. For large motors, it can be set to 5, and for small loads, it can be set to 2 or 3. During the execution, a counter is maintained to record the current continuous length through traversal. If the threshold is exceeded, the starting point of the segment is immediately stored in the trigger point list. Finally, a set of trigger reference point sequences is obtained, such as 50, 120, etc. Then these trigger points are sent to the PLC and converted into a relay action sequence to complete the intelligent adjustment of the trigger timing signal output.

[0113] Please see Figure 5 The specific steps of S4 are as follows:

[0114] S401: Obtain the complete sequence of periodic excitation current of the intelligent adjustment trigger signal, divide it into positive half-cycle current sequence and negative half-cycle current sequence according to the period boundary point, extract the change rate information of the two sequences, and generate positive and negative half-cycle rate sequences.

[0115] The specific calculation formula for extracting the rate of change information of two sequences is as follows:

[0116]

[0117] Calculate the rate characteristic value and generate positive and negative half-cycle rate sequences;

[0118] in, The characteristic value representing the rate of change of the i-th segment of the current sequence. ω represents the current difference (in A) between the j-th sampling point and the previous sampling point in the i-th current sequence. j γ represents the current weighting factor (dimensionless, used to adjust the amplitude) at the j-th sampling point. j The time-varying correction factor (dimensionless, used to characterize time sensitivity) represents the time-varying correction factor for the j-th sampling point. represents the time interval (in seconds) between the j-th sampling point and the previous sampling point in the i-th current sequence, and N represents the total number of sampling points in the i-th current sequence;

[0119] Sampling frequency: 10kHz, sampling interval

[0120] The current value at the j-th sampling point of the i-th current sequence

[0121]

[0122] Current difference

[0123]

[0124] Current weighting factor ω j :

[0125] ω2=1.0, ω3=1.1, ω4=0.9, ω5=0.8;

[0126] Time-varying correction factor γ j :

[0127] γ2=1.0, γ3=1.0, γ4=1.0, γ5=1.0;

[0128] The calculation process is as follows:

[0129] Calculate the rate characteristic component for each sampling point:

[0130] For j=2:

[0131]

[0132] ω2=1.0

[0133] γ2=1.0

[0134]

[0135] For j=3:

[0136]

[0137] ω3=1.1

[0138] γ3=1.0

[0139]

[0140] For j=4:

[0141] ω4=0.9

[0142] γ4 = 1.0

[0143]

[0144] For j=5:

[0145]

[0146] ω5=0.8

[0147] γ5 = 1.0

[0148]

[0149] Calculate the characteristic value of the average rate of change

[0150]

[0151] The results show that the characteristic value of the average rate of change of the current in the i-th segment of the sequence is 3500.000017, in A / s. This value reflects the average rate of change of the current in this segment of the sequence; a larger value indicates a more drastic change in the current. This characteristic value can be used for further analysis of the dynamic characteristics of the current sequence or for applications such as fault detection.

[0152] The formula comprehensively reflects the composite rate of change characteristics of the current sequence by unifying the changes in two different dimensions into a sum of squares followed by a square root. Firstly, the current difference... Multiplying by the current weighting factor ω_j represents the correction effect on the current amplitude at that sampling point, reflecting the weighted modulation of the current amplitude. Its square is used to measure the contribution of this component to the overall rate; secondly, With time-varying correction factor γ j Multiply and then divide by the time interval The quantified value of the instantaneous rate of change is obtained to reflect the abrupt change trend of the current over time. Its square also represents the influence of this part on the overall rate. The sum of the squares of these two parts represents the combined contribution of the two types of changes to the composite rate. They are unified under the same dimension, and the square root is equivalent to taking the Euclidean norm to comprehensively reflect the total rate of change. Finally, the summation of all sampling points and the average value are used to extract the overall average rate of change characteristic value of the current segment. This value contains both amplitude change information and time rate change information, thus completely depicting the dynamic evolution of the sequence.

[0153] The rate characteristic value represents the comprehensive dynamic rate level of current change at each sampling point within the analyzed current sequence range. It not only quantifies the absolute change amplitude of current between different sampling points, but also combines the trend of current change over time. By synthesizing the amplitude change term and the time change rate term in the form of a sum of squares and then averaging them, this characteristic value can simultaneously reflect the intensity and fluctuation sensitivity of the current during that period. It is an important numerical parameter for describing the overall activity of the current sequence. The larger the value, the more significant the current fluctuation and the faster the change.

[0154] S402: Based on the positive and negative half-cycle rate sequences, detect the maximum value in the positive half-cycle rate and the maximum value in the negative half-cycle rate, calculate the difference between the two, and obtain the maximum rate difference.

[0155] Based on the previously obtained positive and negative half-cycle rate sequences, the maximum rate value in the positive half-cycle rate sequence is found by comparing it point by point from beginning to end. For example, when traversing to the 210th data point, the rate reaches 0.42 A / ms, which is higher than all previously compared values, so this value is recorded as the maximum rate of the positive half-cycle. Then, the same method is used to compare all rate values ​​in the negative half-cycle rate sequence point by point. When the rate at the 88th data point reaches 0.31 A / ms, which exceeds all previously compared values, it is recorded as the maximum rate of the negative half-cycle. The maximum rate of the positive half-cycle is then directly subtracted from the maximum rate of the negative half-cycle to obtain the maximum rate difference of 0.11 A / ms. To classify the difference, the following rules can be set: if the maximum rate difference is less than 0.05 A / ms, it is considered a low difference; between 0.05 A / ms and 0.2 A / ms, it is a medium difference; and above 0.2 A / ms, it is a high difference. Thus, 0.11 A / ms can be determined to fall within the medium difference range. In practice, you can also write a simple program, such as using Python, to iterate through all the periodic sequences, calculate the maximum rate difference for each period, and then use conditional statements to classify the difference into the corresponding interval category.

[0156] S403: Based on the maximum rate difference and the triggering state of the intelligent adjustment trigger signal, the condition is determined and filled into the corresponding position of the dynamic sequence. After completing the process, the excitation adjustment dynamic command sequence is obtained.

[0157] Based on the calculated maximum rate difference and the trigger state of the intelligent adjustment trigger signal, a conditional judgment process is executed, assigning values ​​to the dynamic instruction units of each cycle in the sequence sequentially. When the trigger state is valid and the maximum rate difference is within the middle difference range, the dynamic instruction value for that cycle is set to 75; if the maximum rate difference is within the high difference range, it is set to 100; and if it is within the low difference range, it is set to 50. For example, when the trigger state is valid in the 8th cycle and the maximum rate difference is 0.11 A / ms, since it falls within the middle difference range, the corresponding dynamic instruction value for that cycle is set to 75. After each assignment, the value is immediately written to the corresponding dynamic instruction sequence position. All cycles are processed in the same way, and after all cycles are completed, a complete excitation adjustment dynamic instruction sequence is formed, which is periodically arranged into a sequence output.

[0158] Please see Figure 6 The specific steps of S5 are as follows:

[0159] S501: Based on the excitation regulation dynamic command sequence, extract the target regulation reference value and the corresponding time point, and compare the numerical relationship of adjacent reference values ​​in the time point sequence to obtain the reference value change amplitude sequence.

[0160] Based on the extraction of the excitation regulation dynamic command sequence, the historical operating data of the excitation regulator recorded in the power plant control room can be scanned first. The time sequence (e.g., 0, 1, 2, 3, 4, 5, 6, 7, 8, 9s) and the corresponding excitation regulation reference value sequence (e.g., 1.00, 1.05, 1.08, 1.15, 1.12, 1.10, 1.20, 1.25, 1.22, 1.18) can be read. Then, the difference between adjacent reference values ​​is calculated according to the chronological order, resulting in a change amplitude sequence of 0.05, 0.03, 0.07, -0.03, -0.02, 0.10, 0.05, -0.03, -0.04. If 1... When recording the excitation regulation of a 10kV bus tie transformer, similar time series such as 5, 10, 15, 20, 25, 30s may appear, with reference value sequences of 10.1, 10.3, 10.4, 10.7, 10.6, 10.8 respectively. The change amplitude sequence is 0.2, 0.1, 0.3, -0.1, 0.2. Data records are usually stored in the power plant's integrated automation system or background monitoring server in the form of CSV files. They can be directly read in batches by scripts to form a list structure. These differences are regarded as the change amplitude of the reference value in each sampling interval. In addition, they can be extracted in the same way between different regulation command operation cycles to form a series of change amplitude trajectory sequences for subsequent calculations.

[0161] S502: Call the reference value change range sequence and the instantaneous value of the excitation current at the corresponding time point, compare the reference value change range with the excitation current fluctuation range at the time point, select the time point that exceeds the excitation current fluctuation range threshold, and obtain the over-amplitude time index set.

[0162] Once the reference value variation sequence has been obtained, the instantaneous excitation current values ​​under the corresponding time sequence can be further retrieved. For example, if the excitation current sequence is 100, 105, 107, 115, 113, 112, 125, 130, 128, 123 A, then the difference between each segment of this sequence is calculated sequentially according to time, forming a current fluctuation amplitude sequence of 5, 2, 8, -2, -1, 13, 5, -2, -5 A. This sequence is then paired one-to-one with the reference value variation sequence. Each time point forms a triplet recording the time, reference value variation amplitude, and current fluctuation amplitude. Then, by comparing each value, it is determined whether the reference value variation amplitude exceeds a pre-set proportional threshold. The relationship between the reference value variation amplitude and the excitation current fluctuation amplitude can be expressed as 20%. The judgment criterion is that if the absolute value of the change in the reference value exceeds 20% of the current fluctuation amplitude, the time point is selected into the over-amplitude time index set. For example, in segment 7, the reference value change amplitude is 0.05, and the excitation current fluctuation amplitude is 5A. Calculated at 20%, this is 1. Obviously, 0.05 is less than 1, so it is not included in the over-amplitude index. However, when the excitation current fluctuation amplitude is only 1A, if the reference value change is 0.05, it is already greater than 0.2 and is judged as over-amplitude. The threshold can be obtained by multiplying the standard deviation of the short-term current fluctuation amplitude by 20% after statistically analyzing multiple typical operating records. For example, if the short-term fluctuation of the excitation current is around 5A, the threshold range falls around 1A. In this way, the corresponding time point indices can be collected to form an over-amplitude time index set, such as including 3, 7, and 8.

[0163] S503: Locate the corresponding time point of the original target adjustment reference value sequence according to the overamplitude time index set, perform incremental adjustment on the corresponding reference values ​​in sequence, and combine them into a sequence according to the time order to obtain the periodic recursive excitation reference value sequence;

[0164] Once the over-amplitude time index set is obtained, the time points pointed to by these indices can be directly located in the initial reference value sequence. For example, indices 3, 7, and 8 point to times of 3, 7, and 8 seconds, respectively. The reference value changes at these time points are adjusted incrementally for each point. For instance, using 80% as the adjustment coefficient, the change in the third segment (0.07) is adjusted to 0.056. This is then used to correct the reference value for the fourth time point (1.08 + 0.056 = 1.136). The change in the seventh segment (0.05) is then adjusted to 0.04. The reference value for the eighth time point (1.25 + 0.04 = 1.29) is then corrected. Finally, the change in the eighth segment (-0.03) is adjusted... The value is -0.024. The reference value at the 9th time point is then adjusted to 1.196 by subtracting 0.024 from 1.22. After combining all the time points in sequence, a new recursive excitation reference value sequence is obtained: 1.00, 1.05, 1.08, 1.136, 1.12, 1.10, 1.20, 1.29, 1.196, 1.18. This recursive process can be completed by the control script in the power plant automation simulation system. For different generator models, such as those with an excitation rated current range of 300A, this adjustment coefficient can be further refined to 90% or 70% to match different excitation response speed requirements. The data list can also be directly transmitted to the real-time adjustment module for subsequent control applications.

[0165] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for intelligently regulating a generator based on AI, characterized in that, Includes the following steps: S1: Obtain the maximum sampling time point of the excitation current in the positive half-cycle and the minimum sampling time point in the negative half-cycle within the cycle. Use the relationship between these two time points and the cycle length to determine the degree of deviation and obtain the transient excitation deviation judgment conclusion. S2: Based on the transient excitation offset judgment conclusion, the output voltage sequence of the corresponding period is collected, the change amplitude between adjacent sampled values ​​is extracted, and then the condition judgment is performed through the correspondence with the excitation offset conclusion to form the voltage change condition evaluation result; S3: Based on the voltage change condition evaluation results, collect the frequency fluctuation sequence and power factor sequence of the same period, obtain the change rate of adjacent sampling points, and compare the combination relationship between frequency and power change rate with the voltage change condition to obtain the intelligent adjustment trigger signal; S4: Call the intelligent adjustment trigger signal to obtain the maximum rate value of the full sequence of periodic excitation current in the positive and negative half-cycles and perform direct difference calculation. Compare the obtained rate difference with the trigger state of the intelligent adjustment trigger signal to generate a dynamic command sequence for excitation adjustment.

2. The method for intelligently regulating a generator based on AI according to claim 1, characterized in that, The transient excitation offset judgment conclusion includes offset time difference, period ratio coefficient, and offset degree level; the voltage change condition evaluation result includes voltage change amplitude threshold, voltage change trend type, and voltage condition compliance status; the intelligent adjustment trigger signal includes adjustment trigger flag, adjustment execution priority, and trigger validity status; and the excitation adjustment dynamic command sequence includes rate difference reference, command action sequence number, and command applicable period segment.

3. The method for intelligently regulating a generator based on AI according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Obtain the sampling sequence of the excitation current during the positive and negative half-cycles within the cycle. Based on the relationship between the current value and time point of each sampling point in the positive half-cycle, filter the corresponding time point with the largest current value to obtain the peak time point of the positive half-cycle. S102: Call the peak time point of the positive half-cycle, and based on the relationship between the current value and the time point of the sampling point of the negative half-cycle, filter the corresponding time point with the smallest current value, and combine the time point with the interval between the peak time point of the positive half-cycle to obtain the excitation time difference. S103: Based on the ratio of the excitation time difference to the cycle length, and according to the preset offset judgment benchmark, determine the offset degree range and obtain the transient excitation offset judgment conclusion.

4. The method for intelligently regulating a generator based on AI according to claim 3, characterized in that, The specific steps of S2 are as follows: S201: Obtain the transient excitation offset judgment conclusion, collect the corresponding period output voltage sequence, extract the change amplitude of adjacent sample values ​​in the sequence in turn, and generate adjacent sample change sequence; S202: Based on the adjacent sampling change sequence, call the offset amplitude reference value in the excitation offset judgment conclusion, complete the comparison between the change amplitude and the reference value, obtain the proportion exceeding the reference value, and obtain the offset amplitude proportion value; S203: Based on the offset amplitude ratio and the judgment threshold in the excitation offset judgment conclusion, a condition judgment is made to form a normal or abnormal classification result, and the voltage change condition evaluation result is obtained.

5. The method for intelligently regulating a generator based on AI according to claim 4, characterized in that, The specific steps for S3 are as follows: S301: Based on the voltage change condition evaluation results, obtain the frequency fluctuation sequence and power factor sequence of the same period, call the sampling point data of the voltage change condition, deduce the change rate of adjacent sampling points, and generate frequency and power factor change rate sequences. S302: Based on the frequency and power factor change rate sequence, call the voltage change data of the voltage change condition sequence, compare the frequency change rate, power factor change rate and voltage change value in sequence order, filter the points that simultaneously meet the threshold conditions, and obtain the threshold correlation sequence. S303: Based on the threshold association sequence, identify the range of points that continuously meet the conditions, determine the starting position of the range as the trigger reference, and obtain the intelligent adjustment trigger timing signal.

6. The method for intelligently regulating a generator based on AI according to claim 5, characterized in that, The rate of change derived from the sampling interval refers to the degree of change of the sequence over time by dividing the difference in sequence values ​​between two adjacent sampling points by the sampling time interval, given the time interval data of each sampling point. The power factor change rate refers to the time-varying behavior of the power factor sequence, specifically within the derived change rate.

7. The method for intelligently regulating a generator based on AI according to claim 6, characterized in that, The specific steps of S4 are as follows: S401: Obtain the complete sequence of the periodic excitation current of the intelligent adjustment trigger signal, divide it into positive half-cycle current sequence and negative half-cycle current sequence according to the period boundary point, extract the change rate information of the two sequences, and generate positive and negative half-cycle rate sequences. S402: Based on the positive and negative half-cycle rate sequences, detect the maximum value in the positive half-cycle rate and the maximum value in the negative half-cycle rate, calculate the difference between the two, and obtain the maximum rate difference. S403: Based on the maximum rate difference and the triggering state of the intelligent adjustment trigger signal, the condition is determined and filled into the corresponding position of the dynamic sequence. After completing the process, the excitation adjustment dynamic command sequence is obtained.

8. The method for intelligently regulating a generator based on AI according to claim 7, characterized in that, The positive half-cycle current sequence is formed by selecting continuous data segments in the positive direction within the cycle from the obtained full sequence of periodic excitation current based on the cycle boundary point. The negative half-cycle current sequence is also derived from the full cycle excitation current sequence. The continuous data segments with negative current are extracted through the cycle boundary points to form an independent sequence.

9. The method for intelligently regulating a generator based on AI according to claim 1, characterized in that, The method further includes: S5: Based on the excitation adjustment dynamic command sequence, update the adjustment reference value according to the command sequence in combination with the excitation current fluctuation characteristics to generate a periodic recursive excitation reference value sequence. The periodic recursive excitation reference value sequence includes recursive update values, periodic markers, and excitation target values.

10. The method for intelligently regulating a generator based on AI according to claim 9, characterized in that, The specific steps of S5 are as follows: S501: Based on the excitation regulation dynamic command sequence, extract the target regulation reference value and the corresponding time point, and compare the numerical relationship of adjacent reference values ​​in the time point sequence in turn to obtain the reference value change range sequence; S502: Call the reference value change range sequence and the instantaneous value of the excitation current at the corresponding time point, compare the reference value change range with the excitation current fluctuation range at the time point, select the time point that exceeds the excitation current fluctuation range threshold, and obtain the over-amplitude time index set. S503: Locate the corresponding time point of the original target adjustment reference value sequence according to the overamplitude time index set, perform incremental adjustment on the corresponding reference values ​​in sequence, and combine them into a sequence according to the time order to obtain the periodic recursive excitation reference value sequence.

Images