Transformer consumption reduction intelligent control method and system based on power generation end

Abstract

The invention relates to the technical field of industrial big data, in particular to a transformer consumption reduction intelligent control method and system based on a power generation end, and the method comprises the steps: building a source-load space-time correlated industrial big data set, determining an operation reference, and synchronously monitoring real-time electrical parameters; load component decoupling is executed, a basic load component is stripped, and a power fluctuation deviation parameter is extracted; reconstructing equivalent control power of a power generation end based on a nonlinear fluctuation compensation strategy; and constructing a power grid energy efficiency function to execute loss minimization optimization, determining a power supply voltage setting value and driving on-load voltage regulation. Through multi-dimensional data fusion and non-linear load reconstruction, the problem that a traditional method is difficult to deal with random fluctuation of the load is solved, collaborative optimization of transformer loss and line transmission loss is achieved, and operation energy consumption of a power supply loop of a power generation end is effectively reduced.

Description

Technical Field

[0001] This invention relates to the field of industrial big data technology, specifically to a method and system for intelligent control of transformer energy consumption reduction based on the power generation end. Background Technology

[0002] As the core equipment at the source of the power transmission network, the energy efficiency of the power generation transformer directly affects the economics of power generation and grid connection. Currently, energy consumption control of transformers mainly adopts rule-based control technology based on threshold judgment. That is, real-time operating data is collected and a comprehensive score is calculated. Only when the score exceeds a preset static threshold is a switching signal generated to trigger energy-saving operation.

[0003] However, the load environment faced by transformers at the generation end exhibits strong time-varying coupling characteristics. Especially with the increasing proportion of renewable energy grid connection, load data contains both linear trend components based on periodic production schedules and strong nonlinear random fluctuations influenced by weather conditions or sudden dispatching. Existing technologies based on single threshold judgments have significant limitations: they are essentially a "qualitative" decision-making mechanism and cannot quantitatively provide specific voltage regulation target values. When the load is under complex operating conditions with intensely intertwined linear and nonlinear characteristics, simple threshold logic cannot decouple these two characteristics, causing the control strategy to either overreact to small nonlinear fluctuations or lag in responding to long-term linear trend changes, failing to find a dynamic optimal balance between copper and iron losses.

[0004] While general-purpose deep learning algorithms are theoretically capable of handling complex nonlinear data, they face a contradiction between model convergence and generalization ability in industrial big data applications: if undecomposed raw mixed data is directly input into a deep network, the model often struggles to capture long-term linear trends while also taking into account short-term sudden fluctuations, easily getting trapped in local optima, leading to a decrease in load prediction accuracy, and consequently causing the generated energy-saving strategies to deviate from the actual optimal operating point; moreover, relying solely on the "black box" output of deep learning models makes it difficult to construct an interpretable optimization solution space that conforms to physical constraints.

[0005] In summary, the technical problem that existing technologies need to solve is: how to overcome the limitation of single threshold judgment technology in quantitative optimization under the characteristic of strong coupling between linear trend and nonlinear fluctuation in load data, and achieve accurate prediction and quantitative control of transformer operation strategy.

[0006] To address this, a smart control method for reducing transformer power consumption based on the power generation end is proposed. Summary of the Invention

[0007] The purpose of this invention is to provide a transformer loss reduction intelligent control method and system based on the power generation end. By decoupling and analyzing the load components to separate linear trends and random fluctuations, and combining global loss minimization optimization to replace the traditional static threshold judgment, the invention achieves precise quantitative control of transformers under complex coupled loads.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A method and system for intelligent control of transformer power consumption reduction based on the generator end, comprising: Collect generation-side scheduling plans and historical operation records of power supply circuits at the generation end to establish a big data set of industrial power grid with spatiotemporal correlation between source and load, and establish an operation benchmark based on scheduling instructions; synchronously monitor the real-time electrical parameter flow at the generation end, including voltage, current and temperature, through a sensor network; The load component decoupling analysis is performed on the industrial power grid big data set. The rigid benchmark associated with the dispatching command is removed using autoregressive logic as the basic load component. The deviation between the basic load component and the operating benchmark is calculated, and the power fluctuation deviation parameter characterizing unplanned random disturbances is extracted. The power fluctuation deviation parameter is fused with the real-time electrical parameter flow to form a multi-dimensional electrical operating state set. Based on the nonlinear fluctuation compensation strategy, the load correction value is calculated according to the multi-dimensional electrical operating state set and superimposed with the basic load component to reconstruct the equivalent control power at the generation end. A grid energy efficiency function is constructed based on the equivalent control power of the generator end, loss minimization optimization calculation is performed to determine the power supply voltage setting value, and the transformer ratio of the power supply circuit at the generator end is adjusted by on-load tap regulation based on the power supply voltage setting value.

[0009] Preferably, the step of establishing the source-load spatiotemporal correlation industrial power grid big data set specifically includes: accessing the dispatch automation system through an industrial communication interface to read the generation-side dispatch plan, and retrieving historical operation records of the power supply circuit at the generation end from the local monitoring database; mapping the historical operation records to the corresponding geospatial topology nodes based on timestamp indexes, and constructing a multidimensional data structure containing a time-series vector with a time dimension and a topology matrix with a spatial dimension; the industrial power grid big data set includes the following feature fields: generation-side dispatch plan curve, environmental meteorological parameter sequence, historical observation values ​​of transformer winding temperature, and terminal load type labels associated with the power supply circuit at the generation end; the step of establishing the operating benchmark based on dispatch instructions includes: discretizing the generation-side dispatch plan curve to generate a time series with the same time step as the real-time electrical parameter flow, and defining the time series as the operating benchmark; the source-load spatiotemporal correlation is constructed by: using the Pearson correlation coefficient analysis method to quantitatively calculate the hysteresis response parameters between the generation-side dispatch plan and the changes in terminal loads at different spatial nodes, and establishing a spatiotemporal response model.

[0010] Preferably, the step of using autoregressive logic to separate the rigid benchmark associated with the dispatch command as the basic load component specifically includes: establishing a load component decoupling model, defining the load data sequence in the industrial power grid big data set as a linear superposition of deterministic trend components and random fluctuation components; training the industrial power grid big data set using a seasonal autoregressive integral moving average model, inputting the current generation-side dispatch plan as an external regression variable into the trained model, and predicting the theoretical load curve in response to the dispatch command; extracting the continuous interval in the theoretical load curve where the amplitude change rate is lower than a preset steady-state threshold, defining it as the rigid benchmark, and calibrating it as the basic load component.

[0011] Preferably, the step of monitoring the real-time electrical parameters of the generator end, including voltage, current, and temperature, specifically includes: synchronously acquiring instantaneous values ​​of three-phase voltage and three-phase current according to a preset sampling frequency using synchronous phasor measurement units configured on the high-voltage and low-voltage sides of the transformer; monitoring the top oil temperature and winding hot spot temperature of the transformer in real time using fiber Bragg grating sensors; performing signal conditioning and denoising preprocessing on the acquired voltage and current data, wherein the denoising preprocessing includes filtering out noise components using discrete wavelet transform; the power fluctuation deviation parameter characterizing unplanned random disturbances includes: aligning the base load component with the operating reference in the time domain, calculating the amplitude difference between the two at the same sampling moment as the instantaneous deviation sequence; performing statistical feature extraction on the instantaneous deviation sequence, calculating the standard deviation of the instantaneous deviation sequence, and defining the calculation result as the power fluctuation deviation parameter characterizing unplanned random disturbances.

[0012] Preferably, the step of fusing the data into a multi-dimensional electrical operating state set specifically includes: performing multi-source data time registration on the power fluctuation deviation parameter and the real-time electrical parameter stream based on a time synchronization protocol, and using a linear interpolation algorithm to fill the time gap caused by sampling rate differences; performing normalization processing on the registered heterogeneous data, mapping the voltage, current, temperature and power fluctuation deviation parameters to a unified dimensionless interval; and concatenating the normalized parameter data into vectors according to the time step to construct a high-dimensional feature matrix composed of time-series feature dimensions and state feature dimensions, and defining the high-dimensional feature matrix as the multi-dimensional electrical operating state set.

[0013] Preferably, the step of calculating the load correction value and superimposing it with the base load component to reconstruct the equivalent control power at the generation end specifically includes: calculating the time gradient magnitude of the current parameters in the multidimensional electrical operating state set within a sliding time window to characterize the load fluctuation intensity; constructing a nonlinear weighting function, wherein the nonlinear weighting function uses the time gradient magnitude as the independent variable, and the output value range is... The fluctuation weighting coefficient is given, wherein the fluctuation weighting coefficient has a monotonically increasing nonlinear mapping relationship with the time gradient magnitude; the product of the power fluctuation deviation parameter and the fluctuation weighting coefficient is defined as the dynamic loss weighting factor, and the dynamic loss weighting factor is superimposed with the base load component to synthesize the equivalent control power at the generator end that characterizes the additional heating effect of current fluctuation.

[0014] Preferably, the step of constructing the power grid energy efficiency function specifically includes: calculating the transformer internal loss value and the line transmission loss value respectively; the transformer internal loss value consists of an no-load loss component and a load loss component; wherein, the no-load loss component is proportional to the square of the applied voltage, and the load loss component is proportional to the square of the load current; the line transmission loss value is obtained by taking the equivalent control power of the generator end as the transmission load input and performing Joule heating calculation in combination with the equivalent impedance parameters of the power supply circuit at the generator end; introducing the voltage-power sensitivity coefficient of the terminal load, and weighting and summing the transformer internal loss value, the line transmission loss value, and the penalty term based on the sensitivity coefficient to establish a power grid energy efficiency function with the supply voltage as the decision variable.

[0015] Preferably, the step of performing loss minimization optimization calculation to determine the power supply voltage setting value, and adjusting the transformer ratio of the power supply circuit at the generator end through on-load tap changer based on the power supply voltage setting value specifically includes: the loss minimization optimization calculation includes setting a constraint set including the allowable deviation of the power supply voltage, the insulation voltage threshold, and the upper limit of the number of switch actions; under this constraint, the optimal power supply voltage setting value is obtained by iteratively optimizing the power grid energy efficiency function using the particle swarm optimization algorithm; low-pass filtering is performed on the real-time power supply voltage to extract the steady-state trend value; the dead zone range and delay threshold are dynamically corrected based on the power fluctuation deviation parameter, and the delay threshold is limited to not exceeding the maximum allowable duration of voltage over-limit; when the deviation between the voltage steady-state trend value and the setting value exceeds the corrected dead zone and continues to time out, and it is confirmed that the cumulative number of switch actions and the tap position have not reached the limit, the on-load tap changer is driven to perform step-by-step switching so that the deviation falls within the corrected dead zone range.

[0016] A transformer power consumption reduction intelligent control system based on the power generation end includes: Data acquisition and monitoring module: Collects generation-side scheduling plans and historical operation records of power supply circuits at the generation end, establishes a big data set of industrial power grid with spatiotemporal correlation between source and load, and establishes an operating benchmark based on scheduling instructions; synchronously monitors real-time electrical parameters at the generation end, including voltage, current and temperature, through a sensor network; Load decoupling and deviation extraction module: Perform load component decoupling analysis on the industrial power grid big data set, use autoregressive logic to remove the rigid benchmark associated with the dispatching command as the basic load component, calculate the deviation between the basic load component and the operating benchmark, and extract the power fluctuation deviation parameter that characterizes unplanned random disturbances. Multidimensional fusion and load reconfiguration module: The power fluctuation deviation parameter and the real-time electrical parameter flow are fused into a multidimensional electrical operating state set. Based on the nonlinear fluctuation compensation strategy, the load correction value is calculated according to the multidimensional electrical operating state set and superimposed with the basic load component to reconfigure the equivalent control power at the generation end. Energy efficiency optimization and voltage regulation control module: Constructs a grid energy efficiency function based on the equivalent control power of the generator end, performs loss minimization optimization calculation, determines the power supply voltage setting value, and adjusts the transformer ratio of the power supply circuit at the generator end through on-load voltage regulation based on the power supply voltage setting value.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention establishes a large dataset of industrial power grids with spatiotemporal correlation between source and load by collecting generation-side dispatch plans and historical operation records of power supply circuits at the generation end, and establishes an operating baseline based on dispatch commands. This scheme utilizes physical-level dispatch commands as constraints to establish an operating baseline under standard operating conditions from historical data, avoiding the logical biases of purely data-driven methods in the absence of physical constraints. This provides a physically interpretable data foundation for subsequent load analysis and optimization decisions.

[0018] This invention performs load component decoupling analysis on a large industrial power grid dataset, using autoregressive logic to remove the rigid benchmark associated with dispatch instructions as the basic load component. It then calculates the deviation between this component and the operating benchmark, extracting a power fluctuation deviation parameter characterizing unplanned random disturbances. This mechanism separates the coupled deterministic trend component and random fluctuation component in the original load data, reducing mutual interference between different characteristic components during the modeling process, solving the prediction distortion problem caused by mixed load characteristics, and achieving high-precision load sensing.

[0019] This invention integrates power fluctuation deviation parameters with real-time electrical parameter flows into a multi-dimensional electrical operating state set. It then uses a gradient-based nonlinear fluctuation compensation strategy to calculate load correction values, which are then superimposed and reconstructed with the base load component. This strategy adjusts the compensation gain based on the gradient characteristics of the fluctuations. When facing nonlinear disturbances caused by renewable energy access or sudden dispatching, it quantifies the additional Joule heat loss caused by the disturbance and uses the nonlinear fluctuation compensation strategy to find a voltage balance point that minimizes the sum of this additional loss and no-load loss. By dynamically adjusting the voltage reference, the operating impedance characteristics of the power supply circuit at the generator end actively adapt to the current fluctuation conditions, thereby effectively reducing the overall transmission energy consumption during fluctuations.

[0020] This invention constructs a grid energy efficiency function for the power supply circuit at the generator end based on the equivalent control power at the generator end, performs a global loss minimization optimization calculation to determine the power supply voltage setting value, and then regulates the voltage of the power supply circuit at the generator end based on this setting value. This scheme establishes a mathematical model that includes transformer losses and line transmission losses, and can calculate the mathematical equilibrium point between no-load losses and load losses through an optimization algorithm. This replaces static threshold determination, overcomes the limitation of traditional threshold control in quantitative optimization, and reduces the overall operating losses of the power supply circuit at the generator end. Attached Figure Description

[0021] Figure 1 This is a flowchart of an intelligent control method for reducing transformer power consumption based on the power generation end, provided in Embodiment 1 of the present invention. Figure 2 This is a schematic diagram of a transformer energy-saving intelligent control system based on the power generation end, provided in Embodiment 2 of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figures 1 to 2 This invention provides a method and system for intelligent control of transformer power consumption reduction based on the power generation end, the technical solution of which is as follows: A method and system for intelligent control of transformer power consumption reduction based on the generator end, see reference. Figure 1 The specific implementation steps of the present invention include: Collect generation-side scheduling plans and historical operation records of power supply circuits at the generation end to establish a big data set of industrial power grid with spatiotemporal correlation between source and load, and establish an operation benchmark based on scheduling instructions; synchronously monitor the real-time electrical parameter flow at the generation end, including voltage, current and temperature, through a sensor network; The load component decoupling analysis is performed on the industrial power grid big data set. The rigid benchmark associated with the dispatching command is removed using autoregressive logic as the basic load component. The deviation between the basic load component and the operating benchmark is calculated, and the power fluctuation deviation parameter characterizing unplanned random disturbances is extracted. The power fluctuation deviation parameter is fused with the real-time electrical parameter flow to form a multi-dimensional electrical operating state set. Based on the nonlinear fluctuation compensation strategy, the load correction value is calculated according to the multi-dimensional electrical operating state set and superimposed with the basic load component to reconstruct the equivalent control power at the generation end. A grid energy efficiency function is constructed based on the equivalent control power of the generator end, loss minimization optimization calculation is performed to determine the power supply voltage setting value, and the transformer ratio of the power supply circuit at the generator end is adjusted by on-load tap regulation based on the power supply voltage setting value.

[0024] Example 1:

[0025] This embodiment demonstrates the implementation method of the intelligent control method for transformer energy saving based on the generation end provided by the present invention in the dynamic optimization control of transformer energy efficiency in response to industrial load fluctuations at the step-up substation of a thermal power plant. The specific steps are as follows: Furthermore, the steps for establishing the source-load spatiotemporal correlation industrial power grid big data set specifically include: accessing the dispatch automation system through an industrial communication interface to read the generation-side dispatch plan and retrieving historical operation records of the power supply circuit at the generation end from the local monitoring database; mapping the historical operation records to the corresponding geospatial topology nodes based on timestamp indexes to construct a multidimensional data structure containing a time-series vector with a time dimension and a topology matrix with a spatial dimension; the industrial power grid big data set includes the following feature fields: generation-side dispatch plan curve, environmental meteorological parameter sequence, historical observation values ​​of transformer winding temperature, and terminal load type labels associated with the power supply circuit at the generation end; the steps for establishing the operating benchmark based on dispatch instructions include: discretizing the generation-side dispatch plan curve to generate a time series with the same time step as the real-time electrical parameter flow, and defining the time series as the operating benchmark; the source-load spatiotemporal correlation is constructed by: using the Pearson correlation coefficient analysis method to quantitatively calculate the hysteresis response parameters between the generation-side dispatch plan and the changes in terminal loads at different spatial nodes, and establishing a spatiotemporal response model.

[0026] Specifically, regarding data acquisition and structure construction, the system connects to the provincial automatic generation control system via the IEC61850 communication protocol interface to obtain future generation-side dispatch plan curves. These curves are constructed based on the day-ahead generation plan documents and real-time active power control commands issued by the dispatch automation system. The system generates generation-side dispatch plan curves by performing time-series analysis and numerical extraction on the above information, with the time resolution set to a preset duration (e.g., 15 minutes or 1 hour) according to the grid dispatch cycle. The system simultaneously accesses the local monitoring and data acquisition system to retrieve historical operating records of the power supply circuit at the generation end. These records include historical instantaneous values ​​of voltage, current, and active power, as well as synchronously monitored environmental meteorological parameter sequences and historical observations of transformer winding temperatures. Specifically, the environmental meteorological parameter sequences are collected by micro-meteorological sensors deployed at the substation site, including time-series data of ambient temperature, relative humidity, and wind speed. The historical observations of transformer winding temperatures are directly measured by fiber optic sensors embedded in the transformer windings, including hotspot temperature data of the three-phase windings.

[0027] Specifically, when constructing the multidimensional data structure, the processor establishes a topology matrix with the dimension of "number of physical nodes × number of time steps". The system reads the current data at a certain historical moment, encapsulates it into a time-series vector, and fills the vector into the corresponding spatial column position in the topology matrix by querying the mapping relationship table between physical node IDs (such as transformer high-voltage side bushing ID, low-voltage side branch line ID) and matrix column indices. In addition, the system generates terminal load type labels in the industrial power grid big data set based on the electrical characteristics of the power supply circuit at the generator end: when the total harmonic distortion value of the circuit is detected to be higher than the preset standard and the amplitude fluctuation rate value is lower than the preset lower limit, the system marks the data of that period as "stable rectifier load"; when the amplitude fluctuation rate value is detected to be higher than the preset upper limit, the system marks the data of that period as "impact load".

[0028] Specifically, in establishing the operating baseline, the system performs discretization sampling. For situations where the generation-side dispatch plan is a low-frequency stepped curve while the real-time electrical parameter flow is a high-frequency continuous signal, the system uses a linear interpolation algorithm or a zero-order hold algorithm to upsample the low-frequency dispatch plan curve, subdividing the original minute-level time intervals into millisecond-level time steps, ensuring that its sampling frequency matches the real-time electrical parameter flow collected by the sensor network. The system stores the continuous numerical sequence generated after this upsampling process as a time series and uses this sequence as the operating baseline in subsequent control logic.

[0029] Specifically, regarding the establishment of the spatiotemporal response model and the calculation of the hysteresis response parameters, the system executes the following algorithm steps: First, a two-dimensional data mapping table is constructed. The row index of this table corresponds to a unified time step sequence, and the column index corresponds to the identifier of each spatial node. Historical operating data is then filled into the table. Second, the Pearson correlation coefficient analysis method is applied to process the time series data: The system sets a sliding time window covering a preset duration (e.g., 30 to 60 minutes) before and after the data flow, fixes the historical load data column, and controls the generation-side dispatch plan curve to slide point by point along the time axis within this time window. In each sliding step, the processor obtains the current shifted dispatch sequence and the fixed load data sequence, and calculates the Pearson correlation coefficient between these two sequences. The system traverses the entire sliding time window, identifies the time shift corresponding to when the correlation coefficient reaches its maximum value, and records this shift as the hysteresis response parameter of that spatial node. Finally, the system uses this parameter to perform data correction: When generating the operating baseline in real time, the corresponding reverse and forward shift operations are performed on the current dispatch plan data read pointer based on the hysteresis response parameter, and the dispatch plan value pointed to by the pointer after the shift is used as input data.

[0030] This invention establishes a spatiotemporal correlation industrial power grid big data set and uses Pearson correlation coefficient analysis to quantify the lag response parameters between the generation-side dispatch plan and the changes in terminal loads at different spatial nodes, thus establishing a spatiotemporal response model. This setup achieves deep integration of generation-side dispatch commands and local historical operating data, accurately identifying and compensating for the time delay between unit output and terminal load response. It significantly improves the time alignment accuracy of load forecasting and benchmark establishment, and solves the problems of information silos from single data sources and control timing misalignments.

[0031] Furthermore, the step of using autoregressive logic to separate the rigid benchmark associated with the dispatch command as the basic load component specifically includes: establishing a load component decoupling model, defining the load data sequence in the industrial power grid big data set as a linear superposition of deterministic trend components and random fluctuation components; training the industrial power grid big data set using a seasonal autoregressive integral moving average model, inputting the current generation-side dispatch plan as an external regression variable into the trained model to predict the theoretical load curve in response to the dispatch command; extracting the continuous interval in the theoretical load curve where the amplitude change rate is lower than a preset steady-state threshold, defining it as the rigid benchmark, and calibrating it as the basic load component.

[0032] Specifically, the load component decoupling model constructed in this embodiment adopts the algorithm architecture of a seasonal autoregressive integral moving average model. This model includes two parallel computing modules: the first module is an external regression component calculation module, whose input port is connected to the generation-side dispatch plan data stream, using a regression coefficient matrix to map the deterministic trend component determined by the dispatch instructions; the second module is a stochastic time series component calculation module, using autoregressive polynomials and moving average polynomials to capture the autocorrelation characteristics and stochastic fluctuation components in the residual sequence. The system uses the concurrent generation-side dispatch plan data as external regression variables input into the model, and establishes a numerical mapping relationship between dispatch instructions and load trends through model calculations.

[0033] Specifically, the training process for the industrial power grid big data set includes the following steps: First, the system sets the seasonal periodicity parameter of the model based on the ratio of the sampling frequency of the real-time electrical parameter flow to the daily operating cycle (for example, when the sampling interval is 15 minutes, this parameter is set to 96) to match the daily cyclic rhythm in the data. Second, the processor performs a unit root test on the original sequence. If the sequence is non-stationary, the system performs first-order ordinary differencing and seasonal differencing with a step size equal to the seasonal periodicity parameter, until the sequence meets the stationarity requirement. Subsequently, the system sets the grid search range for the autoregressive order and the moving average order, uses the maximum likelihood estimation method to estimate the parameters of the model under different order combinations, and calculates the Akaike Information Content (AIC) criterion value. The processor selects the parameter combination corresponding to the minimum AIC value as the optimal model order and performs a white noise test on the residuals of the trained model to complete the model construction.

[0034] Specifically, in the real-time simulation and benchmark extraction phase, the system inputs the current generation-side dispatch plan into the trained load component decoupling model. The model outputs regression prediction values ​​without random fluctuation terms, i.e., the theoretical load curve. The processor calculates the amplitude change gradient between adjacent sampling points in the theoretical load curve and compares the absolute value of the gradient with a preset steady-state threshold; the preset steady-state threshold is set as a percentage of the transformer's rated capacity (e.g., 3% to 5%). If the calculated absolute value of the gradient is less than the preset steady-state threshold, the processor determines that the current moment is in the steady-state range and retains the data point at that moment as a rigid benchmark; if the absolute value of the gradient exceeds the threshold, the processor determines it as a transient jump and does not retain the data point. Finally, the processor outputs the set of all retained continuous data point sequences as the basic load component.

[0035] This invention utilizes autoregressive logic to establish a decoupled load component model. It employs a seasonal autoregressive integral moving average model combined with external regression variables to separate the load data sequence into a basic load component and a fluctuation component. This method effectively shields the baseline value from high-frequency random noise by extracting the interval where the amplitude change rate is below a preset steady-state threshold. It establishes a stable control reference surface that includes both deterministic trends and eliminates instantaneous disturbances, thus preventing unnecessary adjustment oscillations in the control system caused by chasing minute fluctuations.

[0036] Furthermore, the step of monitoring the real-time electrical parameters of the generator end, including voltage, current, and temperature, specifically includes: synchronously acquiring instantaneous values ​​of three-phase voltage and three-phase current according to a preset sampling frequency using synchronous phasor measurement units configured on the high-voltage and low-voltage sides of the transformer; monitoring the top oil temperature and winding hot spot temperature of the transformer in real time using fiber Bragg grating sensors; performing signal conditioning and denoising preprocessing on the acquired voltage and current data, wherein the denoising preprocessing includes filtering out noise components using discrete wavelet transform; the power fluctuation deviation parameter characterizing unplanned random disturbances includes: aligning the base load component with the operating reference in the time domain, calculating the amplitude difference between the two at the same sampling moment as the instantaneous deviation sequence; performing statistical feature extraction on the instantaneous deviation sequence, calculating the standard deviation of the instantaneous deviation sequence, and defining the calculation result as the power fluctuation deviation parameter characterizing unplanned random disturbances.

[0037] Specifically, regarding hardware deployment and data acquisition, the system deploys synchronous phasor measurement units on the high-voltage side bushing and the low-voltage side outgoing terminals of the transformer. These synchronous phasor measurement units set a preset sampling frequency to a high-frequency value (e.g., 10kHz or higher) that satisfies the Nyquist sampling theorem and can capture transient disturbances. The system reads the instantaneous values ​​of three-phase voltage and three-phase current output by these units, each with an absolute timestamp. For temperature monitoring, for intelligent transformers with built-in sensors, the system directly reads the wavelength drift uploaded by the fiber Bragg grating sensor and converts it into a temperature value. For conventional transformers without pre-installed internal fiber optic sensors, the system collects the top-layer oil temperature data and calls a pre-stored IEC60076-7 standard thermal circuit model algorithm. Using the historical sequence of real-time load current and top-layer oil temperature as input, the system obtains a real-time estimate of the winding hot spot temperature through inversion calculation.

[0038] Specifically, in the signal conditioning and denoising preprocessing stage, the processor first performs amplification and filtering operations on the original voltage and current signals, and then executes the discrete wavelet transform algorithm: the system sets the wavelet basis function (such as the Daubechies wavelet basis function) and the number of decomposition levels, and decomposes the acquired instantaneous signal into approximate coefficients and detail coefficients; the processor applies a soft thresholding or hard thresholding function to truncate the detail coefficients containing high-frequency noise, and retains the signal components that reflect the power frequency and main harmonic characteristics; finally, the system uses the processed coefficients to perform inverse wavelet transform reconstruction and outputs the denoised clean electrical parameter flow.

[0039] Specifically, regarding the algorithm for calculating the power fluctuation deviation parameter, the processor executes the following steps: First, it reads the base load component sequence and the established operating baseline sequence separated in the previous steps, and performs time alignment operations in the time domain based on their timestamp indices to ensure that the data points involved in the calculation correspond to the same physical time. Second, the processor performs a subtraction operation to calculate the difference between the base load component value and the operating baseline value, generating an instantaneous deviation sequence reflecting the degree to which the actual load deviates from the scheduling plan. Finally, the processor performs statistical feature extraction on this sequence: the system sets a sliding time window and calculates the standard deviation value of the instantaneous deviation sequence data within the window; the system outputs this standard deviation value in real time and marks it as the power fluctuation deviation parameter, which serves as the input variable for subsequent energy efficiency function construction and dead-zone dynamic adjustment algorithms.

[0040] In a preferred embodiment, the step of performing statistical feature extraction on the instantaneous deviation sequence is manifested as a complex quantification process of spectral entropy based on short-time Fourier transform. Specifically, this includes performing a short-time Fourier transform on the instantaneous deviation sequence to generate a time-frequency distribution matrix, calculating the Shannon entropy value through power spectral density normalization, and then executing threshold discrimination logic. Only when the spectral entropy value is lower than a preset disorder threshold is it determined to be a structural disturbance, and the reciprocal of the sum of the entropy value and a preset anti-overflow constant is locked as the power fluctuation deviation parameter. When the entropy value is higher than or equal to the disorder threshold, the parameter is set to zero. This utilizes the sensitivity of entropy to signal order to remove background noise and quantify structural impact characteristics. This embodiment leverages the sensitivity of entropy to signal disorder to accurately remove interference from power grid background noise, capturing only structural impacts caused by the start-up and shutdown of large industrial equipment, thereby significantly improving the specificity of the fluctuation parameter in characterizing real load disturbances.

[0041] This invention, by configuring a synchronous phasor measurement unit and a fiber Bragg grating sensor, and combining it with discrete wavelet transform denoising technology, achieves full-dimensional high-frequency monitoring and signal purification of electrical and thermal parameters. By calculating the statistical characteristics of the instantaneous deviation sequence as a power fluctuation deviation parameter, this invention transforms abstract unplanned random disturbances into quantifiable mathematical indicators, providing objective and sensitive data support for the system to determine whether fluctuation compensation needs to be initiated, and significantly improving the system's adaptability to complex electromagnetic environments in industrial settings.

[0042] Furthermore, the step of fusing the data into a multi-dimensional electrical operating state set specifically includes: performing multi-source data time registration on the power fluctuation deviation parameter and the real-time electrical parameter stream based on a time synchronization protocol, and using a linear interpolation algorithm to fill the time gap caused by sampling rate differences; performing normalization processing on the registered heterogeneous data to map the voltage, current, temperature, and power fluctuation deviation parameters to a unified dimensionless interval; and concatenating the normalized parameter data into vectors according to the time step to construct a high-dimensional feature matrix composed of time-series feature dimensions and state feature dimensions, and defining the high-dimensional feature matrix as the multi-dimensional electrical operating state set.

[0043] Specifically, to address the issue of inconsistent sampling rates between the phasor measurement unit and the temperature sensor, the system employs a precision clock synchronization protocol, namely the IEEE 1588 standard protocol, to perform multi-source data time registration. The execution steps of this protocol are as follows: The system sets the phasor measurement unit as the master clock node and the temperature sensor acquisition unit as the slave clock node; the master clock node periodically sends a synchronization message (Sync) and a follow message containing a transmission timestamp; the slave clock node records the message reception timestamp and sends a delay request message; the master clock node records the time of this request reception and replies with a delay response message; the slave clock node calculates the time deviation from the master clock and the network transmission delay based on the four timestamps mentioned above, adjusts the local clock frequency and phase, and achieves time synchronization. Based on this synchronization, the system applies a linear interpolation algorithm to handle timing gaps: for moments on the high-frequency reference axis where voltage and current data exist but direct temperature sampling is missing, the processor reads the adjacent valid temperature measurements before and after that moment, calculates the estimated temperature value for that moment using a linear equation, and fills the estimated value into the corresponding time slot.

[0044] Specifically, in the data standardization phase, the system employs a max-min standardization method to normalize the registered heterogeneous data. The specific calculation logic of this method is as follows: the processor iterates through the numerical sequences of voltage, current, temperature, and power fluctuation deviation parameters within the historical sliding window, identifying the maximum and minimum values ​​of each parameter. For the measured value at the current moment, the processor performs a subtraction operation to calculate the difference between the measured value and the minimum value. Subsequently, the processor calculates the difference between the maximum and minimum values ​​as the range. Finally, the processor divides the aforementioned difference in measured values ​​by this range, using the quotient as the normalized value, thereby mapping parameters with different physical dimensions to a unified dimensionless interval, such as a closed interval from 0 to 1. After normalization, the processor performs vector concatenation: at each unified time step, the system arranges the normalized voltage, current, temperature, and power fluctuation deviation parameter values ​​in a preset order to form a state feature vector, and stacks the vectors from multiple consecutive time steps along the time dimension to construct a high-dimensional feature matrix, outputting a multi-dimensional electrical operating state set.

[0045] This invention, based on a time synchronization protocol and a linear interpolation algorithm, performs time registration and normalization on multi-source heterogeneous data to construct a high-dimensional feature matrix. This measure effectively solves the problem of data timing misalignment caused by inconsistent sampling frequencies of multi-source sensors and eliminates the influence of numerical differences between different physical dimensions. It maps heterogeneous data to a unified feature space, providing standardized, high-quality input data without missing values ​​for subsequent nonlinear fluctuation compensation calculations, thus improving the algorithm's convergence speed and computational stability.

[0046] Furthermore, the step of calculating the load correction value and superimposing it with the base load component to reconstruct the equivalent control power at the generation end specifically includes: calculating the time gradient magnitude of the current parameters in the multidimensional electrical operating state set within a sliding time window to characterize the load fluctuation intensity; constructing a nonlinear weighting function, wherein the nonlinear weighting function uses the time gradient magnitude as the independent variable, and the output value range is... The fluctuation weighting coefficient is given, wherein the fluctuation weighting coefficient has a monotonically increasing nonlinear mapping relationship with the time gradient magnitude; the product of the power fluctuation deviation parameter and the fluctuation weighting coefficient is defined as the dynamic loss weighting factor, and the dynamic loss weighting factor is superimposed with the base load component to synthesize the equivalent control power at the generator end that characterizes the additional heating effect of current fluctuation.

[0047] Specifically, for the quantification of fluctuation intensity, the system first determines the parameters of the sliding time window: the processor reads the sampling time interval of the electrical parameters, divides the preset fluctuation characteristic capture period (e.g., a single power frequency cycle or a millisecond-level transient cycle) by the sampling time interval, and sets the integer part of the calculated quotient as the window length (number of data points); the system constructs a first-in-first-out data queue based on this length to update the data content in real time. Subsequently, the system reads the current parameter sequence within the sliding time window from the multi-dimensional electrical operating state set; the processor performs a first-order difference operation on the current values ​​at adjacent times within the window, calculates the absolute value of the difference between adjacent current values, and marks the average of these differences as the time gradient magnitude. This value is physically used to characterize the intensity of the high-frequency components in the current waveform, because the eddy current loss of the transformer winding and the increase in AC resistance caused by the skin effect are positively correlated with the rate of change of current.

[0048] Specifically, the nonlinear weighting function is a Logistic S-curve function, which the system uses to calculate the fluctuation weight coefficient. The function's operation logic is as follows: the system uses the natural constant e as the base to calculate a power term with the product of a negative slope adjustment factor and the time gradient magnitude deviation (i.e., the gradient magnitude minus its midpoint value). Then, it calculates the reciprocal of the sum of 1 and this power term, outputting the reciprocal as the fluctuation weight coefficient. For parameter calibration, the system uses the sine differential principle to convert the time gradient magnitude into an equivalent frequency and combines this with the skin effect formula to obtain the real-time skin depth. The slope adjustment factor is calculated as follows: the critical gradient magnitude at which the skin depth equals the pre-stored conductor radius is used as the function inflection point, and a sensitivity bandwidth is set according to a certain proportion (e.g., 20%) of this inflection point value to determine the slope. This ensures that when the skin depth caused by the gradient is less than the conductor radius, the weight coefficient rapidly approaches 1, compensating for additional losses at high frequencies.

[0049] Specifically, during the power combining stage, the system calls the stored power conversion coefficient, which is calculated from the transformer's rated voltage and the current line's power factor, to convert the statistical dimension parameter into the physical power dimension. The processor multiplies the statistically calculated power fluctuation deviation parameter (representing the fluctuation energy amplitude) with the power conversion coefficient to obtain the fluctuation equivalent power value in kilowatts. Subsequently, the processor calculates the product of this fluctuation equivalent power value and the fluctuation weighting coefficient (representing the proportion of high-frequency additional losses), and marks this product as the dynamic loss weighting factor. Finally, the system performs an algebraic addition operation on the dynamic loss weighting factor and the base load component, and uses the result as the equivalent control power output at the generator end. This parameter is essentially equivalent to the sum of the Joule heat power generated by the fundamental current and the additional eddy current loss power generated by the high-frequency harmonic current, thus truly reflecting the total thermal effect under complex fluctuation conditions.

[0050] In a preferred embodiment, the step of reconstructing the equivalent control power at the generator end involves establishing a dynamic compensation mechanism based on the fluctuating thermal effect. The processor extracts instantaneous rate-of-change data of the load current from a multi-dimensional electrical operating state set and calculates the moving average of the absolute value of this rate-of-change within a preset sliding window to obtain a load fluctuation intensity index. A pre-stored exponential gain adjustment model is invoked, using the load fluctuation intensity index as an input variable, to calculate and output a dynamic compensation coefficient that increases non-linearly with the fluctuation intensity. The processor performs a multiplication operation on the power fluctuation deviation parameter and the dynamic compensation coefficient to obtain a thermal margin correction amount for unplanned disturbances. Finally, the thermal margin correction amount is algebraically added to the base load component to synthesize the equivalent control power at the generator end, which includes steady-state reference and transient thermal effect compensation. This embodiment, by constructing a dynamic compensation mechanism that increases non-linearly with the load fluctuation intensity, quantifies the additional thermal effect generated by transient disturbances into a power correction amount, enabling the system to proactively reserve a thermal safety margin under severe fluctuation conditions, effectively improving the thermal stability of the transformer to non-linear impact loads.

[0051] This invention constructs a nonlinear weighting function based on the time gradient magnitude, achieving an adaptive response to load fluctuation intensity. By outputting low weights when the load is stable to maintain control stability and high weights when the load fluctuates sharply to quickly compensate for deviations, this invention solves the contradiction between steady-state accuracy and dynamic response that traditional linear compensation cannot simultaneously address. It ensures that the reconstructed equivalent control power at the generator end can accurately reflect the power demand under impact conditions, thereby guiding the transformer to perform more precise voltage regulation.

[0052] Furthermore, the steps for constructing the power grid energy efficiency function specifically include: calculating the transformer internal loss value and the line transmission loss value respectively; the transformer internal loss value consists of an no-load loss component and a load loss component; wherein, the no-load loss component is proportional to the square of the applied voltage, and the load loss component is proportional to the square of the load current; the line transmission loss value is obtained by taking the equivalent control power of the generator end as the transmission load input and performing Joule heating calculation in combination with the equivalent impedance parameters of the power supply circuit at the generator end; introducing the voltage-power sensitivity coefficient of the terminal load, and weighting and summing the transformer internal loss value, the line transmission loss value, and the penalty term based on the sensitivity coefficient to establish a power grid energy efficiency function with the supply voltage as the decision variable.

[0053] Specifically, the calculation method for the transformer internal loss value and the line transmission loss value is as follows: The system first calls the transformer's factory parameters and constructs loss calculation logic accordingly. For the no-load loss component, the processor reads the real-time supply voltage, calculates the square of the ratio of this voltage to the rated voltage, and multiplies this squared value by the rated no-load loss value. For the transformer load loss component and the line transmission loss value, in order to address the influence of load characteristics on the optimization direction, the system performs voltage and power coupling correction. The specific steps are as follows: The processor calls the equivalent control power at the generator end as the reference power, uses the voltage-power sensitivity coefficient as the exponential term, calculates the product of the reference power and the power of the ratio of the real-time supply voltage to the reference voltage, thereby obtaining the corrected power under the current voltage; subsequently, the corrected power is divided by the real-time supply voltage to obtain the equivalent operating current, and the Joule heat calculation is completed based on the product of the square of this current and the impedance parameter, where the impedance parameter includes the real part of the transformer short-circuit impedance and the line resistance value. This step ensures that the model can correctly reflect the actual energy consumption change trend of constant impedance or constant current load during voltage regulation.

[0054] Specifically, the calculation method of the voltage power sensitivity coefficient includes: the system calculates the active power at the generator end in real time based on the real-time electrical parameter flow including voltage and current monitored by the sensor network; the processor uses a preset sliding time window to extract the voltage amplitude sequence and active power sequence within a recent period, and performs natural logarithmic operations on the values ​​in these two sequences to transform the nonlinear exponential relationship into a linear relationship; subsequently, the system uses the least squares algorithm to perform linear regression analysis on the logarithmic active power sequence relative to the logarithmic voltage amplitude sequence, and the slope of the calculated regression line is the voltage power sensitivity coefficient. This coefficient is used to quantitatively characterize the sensitivity of the power consumption of the terminal load to voltage changes under the operating conditions reflected by the current real-time electrical parameter flow.

[0055] Specifically, when constructing the power grid efficiency function, the system utilizes the voltage power sensitivity coefficient and its corresponding transmission load to lock in the waxing and waning patterns of each loss component with voltage changes. The transmission load is the equivalent control power at the generator end obtained from the reconstruction in the previous steps. This power value is composed of the base load component and the dynamic loss weighting factor, used to characterize the comprehensive load level including steady-state trends and transient fluctuation thermal effects. Simultaneously, the system calculates a penalty term based on the sensitivity coefficient to constrain the voltage safety range. The calculation method for the penalty term based on the sensitivity coefficient is as follows: the system presets upper and lower safety thresholds for the supply voltage; if the supply voltage value to be optimized is within the range of these thresholds, the system sets the value of the penalty term to zero; if the supply voltage value exceeds these thresholds, the system calculates the square of the difference between the voltage value and the nearest threshold boundary, and multiplies this square by a preset penalty function constant to obtain the value of the penalty term. Finally, the processor performs a weighted summation operation. Specifically, the method is as follows: the internal loss value of the transformer and the transmission loss value of the line are summed to establish the total physical loss value of the system, and the total physical loss value of the system is summed with the value after a preset penalty weight (e.g., ...). The weighted penalty terms based on the sensitivity coefficient are then linearly superimposed to output the total value of the power grid efficiency function. This function value serves as the fitness value in the particle swarm optimization algorithm, guiding the algorithm to find a physical equilibrium point between transformer iron losses and line copper losses.

[0056] This invention constructs a power grid energy efficiency function that includes transformer internal losses and line transmission losses, transforming the problem of reducing losses in a single device into a system-level global optimization problem. This model quantifies the differentiated impact of voltage regulation on transformer no-load losses (positively correlated with voltage) and line load losses (negatively correlated with voltage), enabling the system to accurately find the voltage balance point with the minimum total loss through mathematical calculations. This avoids the unintended consequences of relying solely on experience in traditional methods, thus maximizing the overall energy efficiency of the power supply circuit at the generation end.

[0057] Further, the step of performing loss minimization optimization calculation to determine the power supply voltage setting value, and adjusting the transformer ratio of the power supply circuit at the generator end through on-load tap changer based on the power supply voltage setting value specifically includes: the loss minimization optimization calculation includes setting a constraint set including the allowable deviation of the power supply voltage, the insulation voltage threshold, and the upper limit of the number of switch actions; under this constraint, the optimal power supply voltage setting value is obtained by iteratively optimizing the power grid energy efficiency function using the particle swarm optimization algorithm; low-pass filtering is performed on the real-time power supply voltage to extract the steady-state trend value; the dead zone range and delay threshold are dynamically corrected based on the power fluctuation deviation parameter, and the delay threshold is limited to not exceeding the maximum allowable duration of voltage over-limit; when the deviation between the steady-state voltage trend value and the setting value exceeds the corrected dead zone and continues to time out, and it is confirmed that the cumulative number of switch actions and the tap position have not reached the limit, the on-load tap changer is driven to perform step-by-step switching so that the deviation falls within the corrected dead zone range.

[0058] Specifically, for the execution of the optimization algorithm, the processor first initializes the multi-dimensional constraint space. The allowable deviation of the supply voltage is set as the percentage fluctuation of the voltage specified by the power grid standard, for example, ±7% of the rated voltage; the insulation voltage threshold is set as the highest effective value of the power frequency voltage that the transformer is allowed to operate for a long time, to prevent over-excitation and damage to the equipment; the constraint set is composed of the above physical boundary conditions. The process of iteratively optimizing the power grid energy efficiency function using the particle swarm optimization algorithm is as follows: the system initializes a group of random particles, each particle representing a potential supply voltage solution; in each iteration, the processor calculates the power grid energy efficiency function value corresponding to each particle as the fitness, and updates the individual optimal position and the group optimal position of the particle; the particle updates its velocity and position vectors according to the current inertia weight, individual cognitive factor, and social learning factor; when the preset number of iterations or fitness convergence accuracy is reached, the system outputs the final group optimal position, and the voltage value corresponding to this position is the optimal supply voltage setting value.

[0059] Specifically, regarding signal processing and adaptive parameter adjustment, the system performs low-pass filtering on the monitored real-time power supply voltage, employing an infinite impulse response digital filter or a moving average filter algorithm, setting the cutoff frequency below the power frequency to remove transient fluctuations and extract the steady-state trend value reflecting the long-term voltage change trend. Simultaneously, the processor establishes a mapping logic between power fluctuation deviation parameters and control parameters: the larger the power fluctuation deviation parameter value, the wider the dead zone range calculated by the system, and the larger the calculated delay threshold; this logic aims to establish a shielding mechanism against high-frequency disturbances. During this process, the processor implements a safety boundary clamping strategy, limiting the delay threshold to no more than the maximum allowable duration of voltage over-limit; the maximum allowable duration of voltage over-limit is a preset upper limit value (e.g., 60 seconds) based on national power quality standards (such as GB / T 12325) and the transformer's short-time overexcitation magnetic withstand characteristic curve. When the calculated adaptive delay threshold exceeds this upper limit, the system forcibly truncates the delay threshold to this upper limit value to prevent the equipment from operating under overvoltage conditions for too long in pursuit of anti-oscillation effects, thereby damaging the insulation.

[0060] Specifically, when an adjustment action is triggered, the system monitors the voltage deviation in real time. When the deviation between the steady-state trend value and the optimal supply voltage setting value exceeds the corrected dead zone range and the duration exceeds the corrected delay threshold, the system performs a dual state check. The specific method is as follows: First, the system reads the switch action records in the memory to confirm that the cumulative number of switch actions has not reached the preset action limit, such as thirty times per day; second, the system reads the position sensor signal to confirm that the current tap position is not at the physical travel limit position, that is, if a voltage increase is required, the current tap position is not the highest; if a voltage decrease is required, the current tap position is not the lowest. Only when all the above conditions are met does the processor send a pulse signal to the motor drive unit to drive the on-load tap changer to perform step-by-step switching, that is, adjusting only one tap at a time, changing the coil turns ratio of the high-voltage side and the low-voltage side of the transformer, so that the steady-state trend value of the supply voltage gradually approaches the allowable deviation range or the optimal setting value of the supply voltage at the physical level.

[0061] Specifically, the core of the control strategy in this embodiment lies in frequency domain separation and thermal balance optimization. The power fluctuation deviation parameter is directly used as an input variable in the preceding grid energy efficiency function calculation. During the global optimization process, the algorithm automatically finds a comprehensive operating point that balances line transmission losses and transformer no-load losses under the current fluctuation intensity. Simultaneously, the system dynamically expands the dead zone range using this power fluctuation deviation parameter: under severe fluctuation conditions, the processor automatically increases the voltage regulation threshold, reducing the system's sensitivity to high-frequency random noise; under stable load conditions, the processor automatically reduces the dead zone range, improving steady-state regulation accuracy. This mechanism limits the on-load tap changer to only perform actions for long-term base load trend changes or persistent energy efficiency deviations, thereby physically mitigating the risk of mechanical oscillation and extending equipment lifespan.

[0062] As a preferred embodiment, the step of adjusting the transformer ratio of the power supply circuit at the generator end specifically includes: introducing an anti-repeated oscillation verification mechanism based on the consistency of trend components; extracting the deterministic trend component obtained from the decoupling of the previous steps and calculating its change gradient in the future preset prediction time domain; identifying the directional attribute of the voltage regulation command output by the loss minimization optimization calculation, wherein the directional attribute is either voltage increase or voltage decrease; if the directional attribute of the voltage regulation command is opposite to the sign direction of the change gradient, and the magnitude of the change gradient exceeds a preset reverse blocking threshold, then the current operating condition is determined to be a transient pseudo-peak interval, the trend blocking logic is activated, and the current on-load tap changer is forcibly kept unchanged until the evolution direction of the deterministic trend component converges with the directional attribute of the voltage regulation command. This embodiment uses macroscopic trend information to filter out short-sighted adjustment behavior caused by local fluctuations, effectively avoiding invalid back-and-forth switching caused by the optimization algorithm over-responding to instantaneous extreme values ​​near the load trend reversal point.

[0063] This invention sets a set of constraints, including voltage deviation, insulation threshold, and number of switching operations, when performing loss minimization optimization calculations, and designs anti-oscillation control logic including dead zone range and delay threshold. This setting ensures that the energy-saving strategy is executed under the premise of meeting equipment safety and lifespan requirements, effectively preventing frequent operation of on-load tap changers caused by instantaneous load spikes. While achieving energy saving and consumption reduction, it maximizes the protection of the mechanical life of the on-load tap changer, achieving a balance between economical operation and equipment reliability.

[0064] Example 2:

[0065] This embodiment demonstrates the application method of the intelligent transformer energy saving control system based on the generator end provided by the present invention in the dynamic optimization control of transformer energy efficiency in response to industrial load fluctuations at the step-up substation of a thermal power plant. (See reference...) Figure 2It includes a data acquisition and monitoring module, a load decoupling and deviation extraction module, a multi-dimensional fusion and load reconfiguration module, and an energy efficiency optimization and voltage regulation control module. The modules communicate and interact with each other through an internal data bus. The specific steps are as follows: Furthermore, the data acquisition and monitoring module is configured as the system's sensing front end, used to construct a large dataset of industrial power grid data with spatiotemporal correlation between source and load and monitor real-time parameters. This module includes an industrial communication interface unit and a sensor network access unit. It connects to the provincial automatic generation control system via the IEC61850 communication protocol interface, reads future generation-side dispatch plan curves, sets their time resolution to a preset duration matching the power grid dispatch cycle, and simultaneously retrieves historical operating records within a preset time period from the plant-level monitoring information system. This module incorporates topology mapping logic, mapping the collected historical operating records to corresponding geospatial topology nodes based on timestamp indices, outputting a multi-dimensional data structure containing a time-series vector and a spatial topology matrix, and setting terminal load type labels based on the electrical characteristics of the power supply circuit at the generation end. Simultaneously, this module, through a physically connected synchronous phasor measurement unit and fiber Bragg grating sensors, synchronously monitors real-time electrical parameter flows including voltage, current, and temperature, and establishes a spatiotemporal response model based on Pearson correlation coefficient analysis, shifting the dispatch plan along the time axis by a corresponding lag time, thereby establishing an operating benchmark based on dispatch commands.

[0066] Furthermore, the load decoupling and deviation extraction module is connected to the data acquisition and monitoring module and is configured to perform deep analysis and separation of load characteristics. This module integrates a seasonal differential autoregressive moving average model and its training engine, using autoregressive logic to separate the load data sequence into deterministic trend components and random fluctuation components. The module uses the current generation-side dispatch plan as an external regression variable to predict the theoretical load curve in response to dispatch instructions, and extracts continuous intervals in the curve where the amplitude change rate is lower than a preset steady-state threshold, labeling them as the base load component. Simultaneously, the module calculates the deviation between the base load component and the spatiotemporally corrected operating baseline, performs statistical feature extraction (such as calculating the standard deviation) on the deviation sequence, extracts the power fluctuation deviation parameter characterizing unplanned random disturbances, and transmits the separated base load component and power fluctuation deviation parameter to the next-level module.

[0067] Furthermore, the multidimensional fusion and load reconfiguration module is connected to the load decoupling and deviation extraction module and the data acquisition and monitoring module, and is configured to generate high-dimensional feature inputs for control decisions and reconfigure real-time loads. This multidimensional fusion and load reconfiguration module first performs multi-source data time registration on the power fluctuation deviation parameter and the real-time electrical parameter stream based on a precise clock synchronization protocol. It then uses a linear interpolation algorithm to fill the time gaps caused by sampling rate differences and performs normalization processing on the registered heterogeneous data to construct a multidimensional electrical operating state set composed of time-series feature dimensions and state feature dimensions. Internally, this module stores a nonlinear weighting function based on the temporal gradient magnitude. This function is set as a monotonically increasing nonlinear mapping relationship; when the gradient magnitude exceeds a preset high threshold and rises, the output weight rapidly approaches 1. The module calculates the load correction value based on this function and performs algebraic superposition with the base load component to synthesize the equivalent control power at the generation end, which includes steady-state reference characteristics and dynamic fluctuation compensation characteristics.

[0068] Furthermore, the energy efficiency optimization and voltage regulation control module is connected to the multi-dimensional fusion and load reconfiguration module, and is configured to calculate the optimal control strategy and drive the on-load tap changer. This module has a built-in power grid energy efficiency function library, which includes calculation models for calculating transformer internal losses (composed of no-load and load loss components) and line transmission losses. Based on the received equivalent control power at the generator end, it constructs the power grid energy efficiency function. This module uses a particle swarm optimization algorithm to perform a global minimum optimization of the power grid energy efficiency function. During the optimization process, it strictly adheres to a preset set of constraints, including the allowable deviation of the supply voltage, the upper limit of transformer insulation tolerance, and the upper limit of the number of switch operations, to determine the supply voltage setting value that minimizes the power grid energy efficiency function value. This module includes hardware driving logic that compares the deviation between the supply voltage in the real-time electrical parameter stream and the supply voltage setting value. When the deviation significantly exceeds the single-stage adjustment range and preset dead zone threshold of the on-load tap changer, and this state continues to exceed the delay threshold, a control command is sent through the hardware interface to drive the motor mechanism of the on-load tap changer to perform the necessary stage switching until the supply voltage returns to the allowable deviation range.

[0069] This invention provides an intelligent control method for transformer energy saving based on the generator end. First, by establishing a large dataset with spatiotemporal correlation between source and load and performing load component decoupling, the complex industrial load is accurately separated into a basic load component associated with dispatch commands and a fluctuation component representing random disturbances. This solves the technical problem of traditional control methods in distinguishing between normal dispatch output and unplanned impact disturbances. Furthermore, by reconstructing the equivalent control power at the generator end through a nonlinear fluctuation compensation strategy, and constructing a grid energy efficiency function accordingly, loss minimization optimization is performed, achieving accurate calculation of the supply voltage setting value. This scheme can dynamically find the optimal balance point between transformer no-load loss and line transmission loss based on real-time load characteristics, and adjust the transformer ratio by driving on-load tap changer. While ensuring power quality, it significantly reduces the overall operating energy consumption of the power supply circuit at the generator end, achieving closed-loop optimization control for transformer operation economy.

[0070] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent control of transformer power consumption reduction based on the generator end, characterized in that, include: Collect generation-side scheduling plans and historical operation records of power supply circuits at the generation end to establish a big data set of industrial power grid with spatiotemporal correlation between source and load, and establish an operation benchmark based on scheduling instructions; synchronously monitor the real-time electrical parameter flow at the generation end, including voltage, current and temperature, through a sensor network; The load component decoupling analysis is performed on the industrial power grid big data set. The rigid benchmark associated with the dispatching command is removed using autoregressive logic as the basic load component. The deviation between the basic load component and the operating benchmark is calculated, and the power fluctuation deviation parameter characterizing unplanned random disturbances is extracted. The power fluctuation deviation parameter is fused with the real-time electrical parameter flow to form a multi-dimensional electrical operating state set. Based on the nonlinear fluctuation compensation strategy, the load correction value is calculated according to the multi-dimensional electrical operating state set and superimposed with the basic load component to reconstruct the equivalent control power at the generation end. A grid energy efficiency function is constructed based on the equivalent control power of the generator end, loss minimization optimization calculation is performed to determine the power supply voltage setting value, and the transformer ratio of the power supply circuit at the generator end is adjusted by on-load tap regulation based on the power supply voltage setting value.

2. The intelligent control method for reducing transformer losses based on the generator end according to claim 1, characterized in that, The steps for establishing the industrial power grid big data set with spatiotemporal correlation between source and load specifically include: The system accesses the dispatch automation system via an industrial communication interface to read the generation-side dispatch plan and retrieves historical operation records of the power supply circuits at the generation end from the local monitoring database. These historical operation records are then mapped to corresponding geospatial topology nodes based on timestamp indexes, constructing a multi-dimensional data structure containing a time-series vector and a spatial topology matrix. The industrial power grid big data set includes the following feature fields: generation-side dispatch plan curve, environmental meteorological parameter sequence, historical observation values ​​of transformer winding temperature, and terminal load type labels associated with the power supply circuits at the generation end. The step of establishing an operating benchmark based on dispatch instructions includes: discretizing the generation-side dispatch plan curve to generate a time series with the same time step as the real-time electrical parameter flow, and defining this time series as the operating benchmark. The source-load spatiotemporal correlation is constructed by: using Pearson correlation coefficient analysis to quantify and calculate the hysteresis response parameters between the generation-side dispatch plan and changes in terminal loads at different spatial nodes, and establishing a spatiotemporal response model.

3. The intelligent control method for reducing transformer losses based on the generator end according to claim 1, characterized in that, The step of using autoregressive logic to strip away the rigid benchmark associated with the scheduling command as the basic load component specifically includes: A load component decoupling model is established, defining the load data sequence in the industrial power grid big data set as a linear superposition of deterministic trend components and random fluctuation components. A seasonal autoregressive integral moving average model is used to train the industrial power grid big data set, and the current generation-side dispatch plan is input into the trained model as an external regression variable to predict the theoretical load curve in response to the dispatch command. The continuous intervals in the theoretical load curve where the amplitude change rate is lower than the preset steady-state threshold are extracted, defined as the rigid benchmark, and calibrated as the basic load component.

4. The intelligent control method for reducing transformer losses based on the generator end according to claim 1, characterized in that, The specific steps for monitoring the real-time electrical parameters of the generator terminal, including voltage, current, and temperature, include: Synchronous phasor measurement units configured on the high-voltage and low-voltage sides of the transformer synchronously acquire instantaneous values ​​of three-phase voltage and three-phase current according to a preset sampling frequency; fiber Bragg grating sensors monitor the transformer top oil temperature and winding hot spot temperature in real time; signal conditioning and denoising preprocessing are performed on the acquired voltage and current data, the denoising preprocessing including filtering out noise components using discrete wavelet transform; the power fluctuation deviation parameter characterizing unplanned random disturbances includes: aligning the base load component with the operating reference in the time domain, calculating the amplitude difference between the two at the same sampling moment as the instantaneous deviation sequence; performing statistical feature extraction on the instantaneous deviation sequence, calculating the standard deviation of the instantaneous deviation sequence, and defining the calculation result as the power fluctuation deviation parameter characterizing unplanned random disturbances.

5. The intelligent control method for reducing transformer losses based on the generator end according to claim 1, characterized in that, The steps of fusing the data into a multi-dimensional electrical operating state set specifically include: Based on the time synchronization protocol, multi-source data time registration is performed on the power fluctuation deviation parameter and the real-time electrical parameter stream. A linear interpolation algorithm is used to fill the time gap caused by the sampling rate difference. Normalization processing is performed on the registered heterogeneous data to map the voltage, current, temperature and power fluctuation deviation parameters to a unified dimensionless interval. The normalized parameter data are vector-concatenated according to the time step to construct a high-dimensional feature matrix composed of time-series feature dimension and state feature dimension, and the high-dimensional feature matrix is ​​defined as the multi-dimensional electrical operating state set.

6. The intelligent control method for reducing transformer losses based on the generator end according to claim 1, characterized in that, The step of calculating the load correction value and superimposing it with the base load component to reconstruct the equivalent control power at the generator end specifically includes: Calculate the time gradient magnitude of the current parameters in the multidimensional electrical operating state set within a sliding time window to characterize the load fluctuation intensity; construct a nonlinear weighting function, which uses the time gradient magnitude as the independent variable, and outputs a value range of [missing value]. The fluctuation weighting coefficient is given, wherein the fluctuation weighting coefficient has a monotonically increasing nonlinear mapping relationship with the time gradient magnitude; the product of the power fluctuation deviation parameter and the fluctuation weighting coefficient is defined as the dynamic loss weighting factor, and the dynamic loss weighting factor is superimposed with the base load component to synthesize the equivalent control power at the generator end that characterizes the additional heating effect of current fluctuation.

7. The intelligent control method for reducing transformer losses based on the generator end according to claim 1, characterized in that, The steps for constructing the power grid energy efficiency function specifically include: The transformer internal loss value and the line transmission loss value are calculated separately. The transformer internal loss value consists of no-load loss component and load loss component. The no-load loss component is proportional to the square of the applied voltage, and the load loss component is proportional to the square of the load current. The line transmission loss value is obtained by taking the equivalent control power of the generator end as the transmission load input and performing Joule heating calculation in combination with the equivalent impedance parameters of the generator end power supply circuit. The voltage and power sensitivity coefficient of the terminal load is introduced, and the transformer internal loss value, the line transmission loss value, and the penalty term based on the sensitivity coefficient are weighted and summed to establish the power grid energy efficiency function with the supply voltage as the decision variable.

8. The intelligent control method for reducing transformer losses based on the generator end according to claim 1, characterized in that, The steps of determining the power supply voltage setting value through loss minimization optimization calculation and adjusting the transformer ratio of the power supply circuit at the generator end through on-load tap regulation based on the power supply voltage setting value specifically include: The loss minimization optimization calculation includes setting a set of constraints, including allowable deviation of supply voltage, insulation voltage threshold, and upper limit of the number of switching operations. Under these constraints, the power grid energy efficiency function is iteratively optimized using a particle swarm optimization algorithm to obtain the optimal supply voltage setting value. Low-pass filtering is performed on the real-time supply voltage to extract the steady-state trend value. The dead zone range and delay threshold are dynamically corrected based on the power fluctuation deviation parameter, and the delay threshold is limited to not exceeding the maximum allowable duration of voltage over-limit. When the deviation between the steady-state voltage trend value and the setting value exceeds the corrected dead zone and continues to exceed the timeout, and it is confirmed that the cumulative number of switching operations and the tap position have not reached the limit, the on-load tap changer is driven to perform step-by-step switching so that the deviation falls within the corrected dead zone range.

9. A transformer loss reduction intelligent control system based on the generator end, characterized in that, include: Data acquisition and monitoring module: Collects generation-side scheduling plans and historical operation records of power supply circuits at the generation end, establishes a big data set of industrial power grid with spatiotemporal correlation between source and load, and establishes an operating benchmark based on scheduling instructions; synchronously monitors real-time electrical parameters at the generation end, including voltage, current and temperature, through a sensor network; Load decoupling and deviation extraction module: Perform load component decoupling analysis on the industrial power grid big data set, use autoregressive logic to remove the rigid benchmark associated with the dispatching command as the basic load component, calculate the deviation between the basic load component and the operating benchmark, and extract the power fluctuation deviation parameter that characterizes unplanned random disturbances. Multidimensional fusion and load reconfiguration module: The power fluctuation deviation parameter and the real-time electrical parameter flow are fused into a multidimensional electrical operating state set. Based on the nonlinear fluctuation compensation strategy, the load correction value is calculated according to the multidimensional electrical operating state set and superimposed with the basic load component to reconfigure the equivalent control power at the generation end. Energy efficiency optimization and voltage regulation control module: Constructs a grid energy efficiency function based on the equivalent control power of the generator end, performs loss minimization optimization calculation, determines the power supply voltage setting value, and adjusts the transformer ratio of the power supply circuit at the generator end through on-load voltage regulation based on the power supply voltage setting value.

Images