Quantile envelope and dynamic characteristic matching additional loss real-time estimation method and system

Abstract

The present application belongs to the technical field of power distribution network energy efficiency analysis, and provides a real-time estimation method and system for distribution transformer additional loss based on quantile envelope and dynamic characteristics. The present application fuses real-time electrical and rated electrical data of the transformer, adopts low quantile point envelope regression technology to accurately fit the benchmark loss envelope model in a strong noise environment, combines a dynamic adjustable exponential characteristic construction mechanism to adaptively learn the optimal nonlinear mapping relationship of the load rate, and continuously tracks the state change of the distribution transformer based on a sliding window retraining trigger mechanism, finally realizes real-time stripping and quantification of the additional loss component, that is, through dynamic calculation of the instantaneous non-negative additional loss and its time cumulative amount, provides high-precision, adaptive and implementable quantitative decision basis for harmonic pollution control and energy saving optimization of core energy-consuming equipment.

Description

Technical Field

[0001] This invention belongs to the field of distribution network energy efficiency analysis technology, and particularly relates to a method and system for real-time estimation of distribution transformer additional losses based on quantile envelope and dynamic characteristics. Background Technology

[0002] The current analysis of distribution transformer losses generally relies on the fixed index model recommended by standards such as IEEE C57.110 and ANSI C57.18. This type of method has three technical limitations: (1) Defects in adaptability to operating conditions: Traditional formulas are based on the assumption of ideal linearity and ignore the complex factors such as load fluctuations in actual operation, such as impact loads, harmonic pollution, and ambient temperature rise, resulting in high errors in the benchmark loss modeling. (2) Invisible additional losses: Existing technologies only output the total loss (no-load loss + load loss) and cannot separate the additional components composed of harmonic eddy current losses and leakage magnetic stray losses, resulting in a lack of accurate targets for energy-saving retrofits. (3) Rigid manual maintenance: The benchmark parameters rely on quarterly manual calibration, such as manually adjusting the coefficients according to the seasonal load curve, which cannot adapt to the real-time evolution of operating conditions, resulting in a high misjudgment rate of losses during extreme winter and summer temperature periods. In summary, fixed models are difficult to construct credible benchmark boundaries in noisy environments, additional losses are in a "black box" state for a long time, and the manual intervention mode lags behind the dynamic needs of the power grid.

[0003] Therefore, there is an urgent need for a benchmark loss envelope modeling technique that integrates multi-source real-time data-driven approaches and has noise immunity capabilities, and to construct a dynamic and adaptive additional loss separation mechanism to overcome the triple bottlenecks of traditional methods in terms of accuracy, real-time performance, and interpretability. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for real-time estimation of additional losses of distribution transformers based on quantile envelope and dynamic characteristics, with the aim of solving the problems mentioned in the background art.

[0005] In a first aspect, the present invention provides a method for real-time estimation of additional losses of distribution transformers based on quantile envelope and dynamic characteristics, comprising the following steps:

[0006] Acquire multi-source data of the transformer, including real-time electrical data and rated electrical data; real-time electrical data includes actual current and actual voltage, and rated electrical data includes rated current, rated voltage, no-load loss and copper loss;

[0007] Calculate the load factor based on the actual current and rated current; calculate the voltage per unit value based on the actual voltage and rated voltage; generate the power-law characteristic of the load factor and the dynamically adjustable exponential characteristic containing a learnable exponent.

[0008] A benchmark loss envelope model is fitted based on a subset of multi-source data with stable operating conditions using quantile regression or deep quantization regression. The benchmark loss envelope model includes a learnable index, which is dynamically optimized by a physical information neural network. The physical information neural network achieves dynamic optimization by constructing a loss function that includes data fitting terms and physical constraint terms.

[0009] Based on the sliding window retraining trigger mechanism, the baseline loss envelope model is adaptively updated.

[0010] Based on the baseline loss envelope model, the non-negative additional loss and its time accumulation are calculated in real time.

[0011] Furthermore, the specific steps are as follows:

[0012] Step S1: Multi-source data acquisition: Real-time acquisition of three-phase current of the distribution transformer at a granularity of at least 15 minutes. and three-phase voltage ;

[0013] Read the transformer's static nameplate parameters, including rated current. Rated voltage No-load loss, copper loss;

[0014] Step S2, Dynamic Feature Construction: Calculate Load Rate The formula is expressed as: per-unit voltage value The formula is expressed as: ;

[0015] Power-law features of the generated load rate , and includes learnable index Dynamically adjustable exponential characteristics ;

[0016] Step S3, Adaptive Quantile Envelope Fitting: By employing quantile regression or deep quantization regression with a quantile parameter of 0.05, a baseline loss envelope model is fitted based on a subset of multi-source data with load rate fluctuation <5%, voltage harmonic distortion rate <2%, and ambient temperature standard deviation <2℃. The formula is expressed as:

[0017] ;

[0018] In the formula, This is the predicted loss value; The regression coefficients characterize the baseline value of no-load loss; The regression coefficients characterizing the variable copper loss intensity; The regression coefficients characterize the effect of voltage on losses;

[0019] Among them, the learnable index Dynamic optimization is achieved through a physical information neural network, which constructs a model containing data fitting terms. With physical constraints The loss function is used to achieve dynamic optimization;

[0020] In the formula, Copper loss coefficient; This represents the actual total loss; The Euclidean norm is used to quantify the error between the actual total loss and the predicted loss. The rate at which losses change with load rate; The norm is used to quantify bias.

[0021] When either condition A or B is met, retraining is triggered to achieve adaptive updating of the baseline loss envelope model:

[0022] A. The sliding window period is 30 days;

[0023] B. The mean residual growth of the sample set is >10%;

[0024] The sample set consists of data collected at multiple time points within a corresponding time period, with each data point including load rate. per-unit voltage and actual total loss .

[0025] Step S4: Real-time estimation of additional losses: Calculate the instantaneous additional losses in real time based on the baseline loss envelope model. The formula is expressed as:

[0026] ;

[0027] In the formula, This is a function that maximizes the value of the additional loss, used to ensure that the additional loss is non-negative. The actual total loss at time t; The loss prediction value at time t;

[0028] Output instantaneous additional loss, daily cumulative additional loss, and monthly cumulative additional loss.

[0029] Furthermore, in step S1, multi-source data acquisition also includes: collecting ambient air temperature, oil temperature, or cooling method;

[0030] Cooling method participates in threshold adjustment:

[0031] ONAN cooling method: Oil temperature > 70℃ triggers sliding window periodic adjustment;

[0032] OFAN cooling method: Maintain the original threshold;

[0033] ODAF cooling method: Oil temperature > 85℃ triggers sliding window period adjustment.

[0034] Furthermore, in step S3, the conditions for triggering retraining also include cooling system alarm linkage: oil temperature >70℃ when ONAN is cooling or oil temperature >85℃ when ODAF is cooling.

[0035] Furthermore, in step S3, the sliding window period is dynamically adjusted according to seasonal load changes:

[0036] When the oil temperature is >75℃ or the ambient temperature is >35℃, the sliding window cycle is 15 days.

[0037] When the monthly change in load factor is greater than 15%, the sliding window period is 30 days.

[0038] When the ambient temperature is below 5℃ for 7 consecutive days, the sliding window period is 45 days.

[0039] Furthermore, in step S3, the sample set undergoes dynamic management training:

[0040] Initial admission criteria: load rate fluctuation <5%, voltage harmonic distortion rate <2%, ambient temperature standard deviation <2℃;

[0041] Dynamic weights: Calculating sample weights The formula is expressed as:

[0042] ;

[0043] In the formula, For the first The actual total measurement loss corresponding to each sample; For the first The baseline loss envelope model prediction value for each sample; For the first The actual total measurement loss corresponding to each sample; For the first The benchmark loss envelope model prediction value for each sample This is the absolute value of the predicted residual;

[0044] Elimination mechanism: Calculate the relative residual value of the sample. The formula is expressed as:

[0045] ;

[0046] When the relative value of the sample residuals If the percentage is greater than 15% for three consecutive times, the sample will be removed from the sample set.

[0047] Secondly, the present invention provides a real-time estimation system for additional losses of distribution transformers based on quantile envelope and dynamic characteristics, comprising:

[0048] Multi-source acquisition module: used to acquire multi-source data of the transformer, including actual current, actual voltage and rated electrical quantities;

[0049] Dynamic Feature Engine: Used to calculate the load rate based on the actual current and rated current; calculate the voltage per unit value based on the actual voltage and rated voltage; generate the power feature of the load rate and the dynamically adjustable exponential feature containing a learnable exponent. The learnable exponent is dynamically optimized through the physical information neural network module, which is integrated into the adaptive envelope fitting module.

[0050] Adaptive envelope fitting module: used to fit a benchmark loss envelope model based on a stable subset of multi-source data through quantile regression or deep quantization regression;

[0051] Additional loss calculation module: used to calculate non-negative additional loss and its time accumulation in real time based on the baseline loss envelope model.

[0052] Thirdly, the present invention provides a computer device, comprising: one or more processors; the processors being used to store one or more programs; and when the one or more programs are executed by the one or more processors, implementing a method for real-time estimation of distribution transformer additional losses based on quantile envelope and dynamic characteristics.

[0053] Fourthly, the present invention provides a computer-readable storage medium, characterized in that it stores a computer program thereon, which, when executed, implements a method for real-time estimation of additional losses of distribution transformers based on quantile envelope and dynamic characteristics.

[0054] The present invention has the following beneficial effects:

[0055] (1) By integrating the real-time electrical and rated electrical data of the transformer, the low quantile envelope regression technology is used to accurately fit the benchmark loss envelope model in a strong noise environment. Combined with the dynamic adjustable exponential feature construction mechanism, the optimal nonlinear mapping relationship of the load rate is adaptively learned. Based on the sliding window retraining trigger mechanism, the state changes of the distribution transformer are continuously tracked. Finally, the additional loss component is stripped and quantified in real time. That is, by dynamically calculating the instantaneous non-negative additional loss and its time accumulation, a high-precision, adaptive, and feasible quantitative decision-making basis is provided for the harmonic pollution control and energy-saving optimization of the core energy-consuming equipment (distribution transformer).

[0056] (2) By using an adaptive quantization envelope mechanism and a dynamic feature construction engine, the additional loss component of the transformer can be accurately separated in a strong noise environment. Compared with the traditional fixed formula scheme, the baseline loss fitting error is reduced by more than 70%, and the additional loss can be quantified and separated by 100%. Combined with the sliding window retraining trigger mechanism (30-day cycle adaptive update), the equipment load and temperature evolution are dynamically tracked, which completely solves the industry problem of seasonal deviation of more than 35% caused by seasonal fluctuations in the traditional method. Attached Figure Description

[0057] Exemplary embodiments of the present invention can be more fully understood by referring to the following figures:

[0058] Figure 1 The flowchart illustrates a method for real-time estimation of transformer additional losses based on quantile envelope and dynamic characteristics, as provided in an embodiment of the present invention.

[0059] Figure 2 This is a schematic diagram of the adaptive quantile envelope fitting process provided in an embodiment of the present invention.

[0060] Figure 3 The flowchart illustrates a real-time estimation system for additional transformer losses based on quantile envelope and dynamic characteristics, as provided in an embodiment of the present invention. Detailed Implementation

[0061] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0062] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention.

[0063] like Figure 1 As shown, this embodiment of the invention provides a real-time estimation method for additional transformer losses based on quantile envelope and dynamic characteristics, including the following steps:

[0064] Acquire multi-source data of the transformer, including real-time electrical data and rated electrical data; real-time electrical data includes actual current and actual voltage, and rated electrical data includes rated current, rated voltage, no-load loss and copper loss;

[0065] Calculate the load factor based on the actual current and rated current; calculate the voltage per unit value based on the actual voltage and rated voltage; generate the power-law characteristic of the load factor and the dynamically adjustable exponential characteristic containing a learnable exponent.

[0066] A benchmark loss envelope model is fitted based on a subset of multi-source data with stable operating conditions using quantile regression or deep quantization regression. The benchmark loss envelope model includes a learnable index, which is dynamically optimized by a physical information neural network. The physical information neural network achieves dynamic optimization by constructing a loss function that includes data fitting terms and physical constraint terms.

[0067] Based on the sliding window retraining trigger mechanism, the baseline loss envelope model is adaptively updated.

[0068] Based on the baseline loss envelope model, the non-negative additional loss and its time accumulation are calculated in real time.

[0069] In some embodiments, the specific steps are as follows:

[0070] Step S1: Multi-source data acquisition: Real-time acquisition of three-phase current of the distribution transformer at a granularity of at least 15 minutes. and three-phase voltage ;

[0071] Read the transformer's static nameplate parameters, including rated current. Rated voltage No-load loss, copper loss;

[0072] Specifically, multi-source data acquisition may also include: collecting ambient air temperature, oil temperature, or cooling method;

[0073] Cooling method participates in threshold adjustment:

[0074] ONAN cooling method: Oil temperature > 70℃ triggers sliding window periodic adjustment;

[0075] OFAN cooling method: Maintain the original threshold;

[0076] ODAF cooling method: Oil temperature > 85℃ triggers sliding window periodic adjustment;

[0077] Specifically, real-time acquisition of three-phase current of distribution transformers ( ) and three-phase voltage ( The data granularity is 15 minutes; ambient air temperature, oil temperature and cooling method (ONAN / OFAN / ODAF) are collected simultaneously; three-phase current and three-phase voltage need to be filtered by sliding window to suppress noise, and the voltage signal needs to be analyzed for total harmonic distortion (THD). When the THD calculated by harmonic analysis is >2%, the data for that period is marked as invalid.

[0078] Obtain the rated current from the transformer nameplate. ), rated voltage ( Core parameters such as no-load loss and copper loss are included; if the rated value is missing, it is supplemented by reverse calculation using the 95% confidence interval of the historical load rate distribution; the copper loss coefficient is preset according to the transformer model. ) and initial learnable index ( For example, the default value of the S11 type transformer. ;

[0079] Step S2, Dynamic Feature Construction: Calculate Load Rate The formula is expressed as: per-unit voltage value The formula is expressed as: ;

[0080] Specifically, calculate the load factor. ;

[0081] per-unit voltage ;

[0082] In the formula, This refers to the instantaneous phase A current; This refers to the instantaneous phase B current; This refers to the instantaneous C-phase current; This refers to the voltage between phase A and phase B. This refers to the voltage between phase B and phase C. This refers to the voltage between phase A and phase C.

[0083] Power-law features of the generated load rate (Characteristics of the main copper loss term) (Characterizing stray loss components) and including learnable exponents Dynamically adjustable exponential characteristics ;

[0084] like Figure 2 As shown, step S3, adaptive quantile envelope fitting: By using quantile regression or deep quantization regression (DQR) with a quantile parameter of 0.05 (τ=0.05), based on the load factor fluctuation <5%, that is, within 6 consecutive sampling points... (Load rate change rate / average load rate < 0.5), voltage harmonic distortion rate < 2%, ambient temperature standard deviation < 2℃. A multi-source data subset fitting reference loss envelope model is given by the following formula:

[0085] ;

[0086] In the formula, This is the predicted loss value; The regression coefficients characterize the baseline value of no-load loss; The regression coefficients characterizing the variable copper loss intensity; The regression coefficients characterize the effect of voltage on losses;

[0087] Among them, the learnable index Through dynamic optimization using a Physical Information Neural Network (PINN), the PINN constructs a model containing data fitting terms. With physical constraints The loss function is used to achieve dynamic optimization;

[0088] In the formula, Copper loss coefficient; This represents the actual total loss; The Euclidean norm is used to quantify the error between the actual total loss and the predicted loss. The rate at which losses change with load rate; The norm is used to quantify bias.

[0089] When either condition A or B is met, retraining is triggered to achieve adaptive updating of the baseline loss envelope model:

[0090] A. The sliding window cycle is 30 days (i.e., the cumulative running cycle reaches 30 days, and it updates naturally).

[0091] B. The mean growth of the sample set residuals is >10% (anomaly detection);

[0092] The sample set consists of data collected at multiple time points within a corresponding time period, with each data point including load rate. per-unit voltage and the corresponding actual total loss ;

[0093] Specifically, the conditions for triggering retraining also include meeting C: cooling system alarm linkage (oil temperature >70℃ when ONAN is cooling or oil temperature >85℃ when ODAF is cooling).

[0094] The sliding window period is dynamically adjusted according to seasonal load changes:

[0095] When the oil temperature is >75℃ or the ambient temperature is >35℃, the sliding window cycle is 15 days.

[0096] When the monthly change in load factor is greater than 15%, the sliding window period is 30 days.

[0097] When the ambient temperature is below 5℃ for 7 consecutive days, the sliding window period is 45 days.

[0098] Step S4: Real-time estimation of additional losses: Calculate the instantaneous additional losses in real time based on the baseline loss envelope model. The formula is expressed as:

[0099] ;

[0100] In the formula, This is a function that maximizes the value of the additional loss, used to ensure that the additional loss is non-negative. The actual total loss at time t; The loss prediction value at time t;

[0101] Output instantaneous additional loss Daily cumulative additional losses and monthly accumulated additional losses ;

[0102] Specifically, the daily cumulative additional loss is the sum of the additional losses from the 96 sampling points (4 times / hour) on that day.

[0103] Monthly cumulative additional losses: sum of the cumulative values ​​for each day of the month;

[0104] Output control: Real-time output of instantaneous additional loss value (15-minute granularity), daily cumulative report generated at 00:05, monthly cumulative report generated at 00:10 on the first day of each month, data is transmitted synchronously via JSON format and SQL database.

[0105] This embodiment achieves accurate quantification of the additional losses of distribution transformers through four major steps: multi-source data acquisition, dynamic feature construction, adaptive quantile envelope fitting, and real-time estimation of additional losses. First, three-phase current and voltage, ambient air temperature, oil temperature, and cooling method are simultaneously collected at a 15-minute granularity. A static benchmark is established by combining this data with transformer nameplate parameters (rated current and voltage, no-load loss, and copper loss). A multi-source data subset with voltage harmonic distortion rate <2%, load rate fluctuation <5%, and ambient temperature standard deviation <2% is rigorously selected for modeling. Second, the load rate and voltage per-unit value are calculated in real time to construct a power-law feature set of the load rate. Then, a 0.05 quantile regression or deep quantization regression is used to fit the benchmark loss envelope model, and the learnable index is dynamically optimized through a physical information neural network. Its loss function integrates data fitting terms and physical constraint terms. Retraining of the baseline loss envelope model is triggered when any of the following conditions are met: a 30-day period, a residual mean growth >10%, or a cooling system threshold. Finally, non-negative additional losses are calculated in real time based on the baseline loss envelope model, outputting instantaneous, daily, and monthly cumulative values ​​of the additional losses. Dynamic weighting and residual elimination mechanisms ensure model robustness. This invention uses quantile envelope separation of loss benchmarks, dynamic optimization of learnable exponents to integrate electromagnetic constraints, and linkage with cooling conditions to achieve adaptive thermal stress modeling, providing core technical support for abnormal loss diagnosis in distribution networks.

[0106] In some embodiments, in step S3, the sample set is dynamically managed for training:

[0107] Initial admission criteria: load rate fluctuation <5%, voltage harmonic distortion rate <2%, ambient temperature standard deviation <2℃;

[0108] Dynamic weights: Calculating sample weights The formula is expressed as:

[0109] ;

[0110] In the formula, For the first The actual total measurement loss corresponding to each sample; For the first The baseline loss envelope model prediction value for each sample; For the first The actual total measurement loss corresponding to each sample; For the first The benchmark loss envelope model prediction value for each sample This is the absolute value of the predicted residual;

[0111] Elimination mechanism: Calculate the relative residual value of the sample. The formula is expressed as:

[0112] ;

[0113] When the relative value of the sample residuals If the percentage is greater than 15% for three consecutive times, the sample will be removed from the sample set.

[0114] In some embodiments, such as Figure 3 As shown, this embodiment of the invention provides a real-time estimation system for additional losses of distribution transformers based on quantile envelope and dynamic characteristics, including:

[0115] Multi-source acquisition module: used to acquire multi-source data of the transformer, including actual current, actual voltage and rated electrical quantities;

[0116] Dynamic Feature Engine: Used to calculate the load rate based on the actual current and rated current; calculate the voltage per unit value based on the actual voltage and rated voltage; generate the power feature of the load rate and the dynamically adjustable exponential feature containing a learnable exponent. The learnable exponent is dynamically optimized through the physical information neural network module, which is integrated into the adaptive envelope fitting module.

[0117] Adaptive envelope fitting module: used to fit a benchmark loss envelope model based on a stable subset of multi-source data through quantile regression or deep quantization regression;

[0118] Additional loss calculation module: used to calculate non-negative additional loss and its time accumulation in real time based on the baseline loss envelope model.

[0119] In some embodiments, the present invention provides a computer device, including: one or more processors; the processors are used to store one or more programs; when the one or more programs are executed by the one or more processors, a method for real-time estimation of transformer additional losses based on quantile envelope and dynamic characteristics is implemented.

[0120] In some embodiments, the present invention provides a computer-readable storage medium, characterized in that it stores a computer program thereon, which, when executed, implements a method for real-time estimation of additional losses of distribution transformers based on quantile envelope and dynamic characteristics.

[0121] Example 1:

[0122] 1. Implementation Environment Configuration

[0123] Transformer model: S11-M-400 / 10 (rated capacity 400kVA);

[0124] Cooling method: ONAN (oil-immersed self-cooling);

[0125] Data acquisition terminal: Supports 15-minute three-phase current / voltage synchronous acquisition, and integrates temperature sensors (ambient temperature, oil temperature);

[0126] Nameplate parameters:

[0127] Rated current Rated voltage ;

[0128] No-load loss copper loss ;

[0129] 2. Specific Implementation Steps

[0130] (1) Multi-source data acquisition

[0131] Real-time acquisition of 15-minute three-phase current ( ) and line voltage ( );

[0132] Simultaneous acquisition of ambient temperature and oil temperature ;

[0133] Read the static nameplate parameters of the transformer: Rated current Rated voltage No-load loss and copper loss ;

[0134] (2) Dynamic feature construction

[0135] Calculate load rate and voltage per unit value:

[0136] ;

[0137] Generate power-law features: (Main item of copper loss) (stray loss component) (Dynamic exponential characteristics, initial p=1.8);

[0138] (3) Adaptive quantile envelope fitting

[0139] Training set selection:

[0140] Select a subset of data with load rate fluctuation <5%, voltage harmonic distortion rate <2%, and ambient temperature standard deviation <2°C;

[0141] Quantile regression modeling (τ=0.05):

[0142] ;

[0143] Physical Information Neural Network (PINN) dynamically optimizes the learnable index Its loss function incorporates data fitting terms and physical constraints (k=0.021 is the copper loss coefficient);

[0144] Retraining trigger conditions:

[0145] The cycle lasts up to 30 days (natural renewal);

[0146] The mean growth of the sample residuals is greater than 10%;

[0147] Oil temperature >70℃ (ONAN cooling method threshold);

[0148] (4) Real-time estimation of additional losses

[0149] Instantaneous additional loss calculation:

[0150]

[0151] Cumulative loss statistics:

[0152] Daily cumulative: summation of 96 sampling points (4 times / hour, 24 hours);

[0153] Monthly cumulative: The sum of the cumulative values ​​for each day.

[0154] Output control:

[0155] Output instantaneous values ​​in real time (JSON format) and generate daily / monthly cumulative reports (SQL storage).

[0156] 3. Example of dynamic parameter adjustment

[0157] Scenario: Summer ambient temperature consistently >35°C, with monthly load rate variation reaching 18%; Adaptive Response:

[0158] When the retraining condition is triggered (load rate change > 15%), the sliding window period is shortened to 30 days (originally 45 days).

[0159] Sample Management:

[0160] Relative value of elimination residual Samples with a percentage greater than 15% for three consecutive times.

[0161] 4. Verification of technical effectiveness

[0162] Reference loss error: Measured error 3.1% (target ≤ 3.2%);

[0163] Additional loss separation rate: 100% quantization output;

[0164] Energy saving guidance value: Monthly cumulative additional losses matched with harmonic mitigation achieve energy savings of up to 95.7%.

[0165] This embodiment 1 achieves high-precision modeling of the baseline loss and real-time stripping of additional losses by integrating multi-source data acquisition, dynamic feature construction, and adopting adaptive quantized envelope technology. The method uses a 30-day sliding window retraining mechanism to ensure adaptive updates and finally outputs the instantaneous value of non-negative additional losses. This overcomes the problems of poor adaptability to operating conditions and invisible additional losses in traditional fixed formulas, reduces the baseline loss error to 3.2%, achieves a 100% quantification rate for additional loss separation, and achieves an energy-saving matching rate of over 95% for monthly cumulative amount-guided harmonic management, significantly improving the level of lean energy efficiency management of the power grid.

[0166] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for real-time estimation of additional losses of distribution transformers based on quantile envelope and dynamic characteristics, characterized in that, Includes the following steps: Step S1: Multi-source data acquisition: Real-time acquisition of three-phase current of the distribution transformer at a granularity of at least 15 minutes. and three-phase voltage ; Read the transformer's static nameplate parameters, including rated current. Rated voltage No-load loss, copper loss; Step S2, Dynamic Feature Construction: Calculate Load Rate The formula is expressed as: per-unit voltage value The formula is expressed as: ; Power-law features of the generated load rate , and includes learnable index Dynamically adjustable exponential characteristics ; Step S3, Adaptive Quantile Envelope Fitting: By employing quantile regression or deep quantization regression with a quantile parameter of 0.05, a baseline loss envelope model is fitted based on a subset of multi-source data with load rate fluctuation <5%, voltage harmonic distortion rate <2%, and ambient temperature standard deviation <2℃. The formula is expressed as: ； In the formula, This is the predicted loss value; The regression coefficients characterize the baseline value of no-load loss; The regression coefficients characterizing the variable copper loss intensity; The regression coefficients characterize the effect of voltage on losses; Among them, the learnable index Dynamic optimization is achieved through a physical information neural network, which constructs a model containing data fitting terms. With physical constraints The loss function is used to achieve dynamic optimization; In the formula, Copper loss coefficient; This represents the actual total loss; The Euclidean norm is used to quantify the error between the actual total loss and the predicted loss. The rate at which losses change with load rate; The norm is used to quantify bias. When either condition A or B is met, retraining is triggered to achieve adaptive updating of the baseline loss envelope model: A. The sliding window period is 30 days; B. The mean residual growth of the sample set is >10%; The sample set consists of data collected at multiple time points within a corresponding time period, with each data point including load rate. per-unit voltage and actual total loss ; Step S4: Real-time estimation of additional losses: Calculate the instantaneous additional losses in real time based on the baseline loss envelope model. The formula is expressed as: ； In the formula, This is a function that maximizes the value of the additional loss, used to ensure that the additional loss is non-negative. The actual total loss at time t; The loss prediction value at time t; Output instantaneous additional loss, daily cumulative additional loss, and monthly cumulative additional loss.

2. The method for real-time estimation of transformer additional losses based on quantile envelope and dynamic characteristics as described in claim 1, characterized in that: In step S1, multi-source data acquisition also includes: collecting ambient air temperature, oil temperature, or cooling method; Cooling method participates in threshold adjustment: ONAN cooling method: Oil temperature > 70℃ triggers sliding window periodic adjustment; OFAN cooling method: Maintain the original threshold; ODAF cooling method: Oil temperature > 85℃ triggers sliding window period adjustment.

3. The method for real-time estimation of transformer additional loss based on quantile envelope and dynamic characteristics as described in claim 2, characterized in that: In step S3, the conditions for triggering retraining also include cooling system alarm linkage: oil temperature >70℃ when ONAN is cooling or oil temperature >85℃ when ODAF is cooling.

4. The method for real-time estimation of additional losses of distribution transformers based on quantile envelope and dynamic characteristics as described in claim 3, characterized in that: In step S3, the sliding window period is dynamically adjusted according to seasonal load changes: When the oil temperature is >75℃ or the ambient temperature is >35℃, the sliding window cycle is 15 days. When the monthly change in load factor is greater than 15%, the sliding window period is 30 days. When the ambient temperature is below 5℃ for 7 consecutive days, the sliding window period is 45 days.

5. The method for real-time estimation of transformer additional loss based on quantile envelope and dynamic characteristics as described in claim 4, characterized in that: In step S3, the sample set is dynamically managed for training: Initial admission criteria: load rate fluctuation <5%, voltage harmonic distortion rate <2%, ambient temperature standard deviation <2℃; Dynamic weights: Calculating sample weights The formula is expressed as: ； In the formula, For the first The actual total measurement loss corresponding to each sample; For the first The baseline loss envelope model prediction value for each sample; For the first The actual total measurement loss corresponding to each sample; For the first The benchmark loss envelope model prediction value for each sample This is the absolute value of the predicted residual; Elimination mechanism: Calculate the relative residual value of the sample. The formula is expressed as: ； When the relative value of the sample residuals If the percentage is greater than 15% for three consecutive times, the sample will be removed from the sample set.

6. A real-time estimation system for additional losses of distribution transformers based on quantile envelope and dynamic characteristics, characterized in that: The system implements the real-time estimation method for additional losses of distribution transformers based on quantile envelope and dynamic characteristics as described in any one of claims 1-5. The system includes: Multi-source acquisition module: used to acquire multi-source data of the transformer, including actual current, actual voltage and rated electrical quantities; Dynamic Feature Engine: Used to calculate the load rate based on the actual current and rated current; calculate the voltage per unit value based on the actual voltage and rated voltage; generate the power feature of the load rate and the dynamically adjustable exponential feature containing a learnable exponent. The learnable exponent is dynamically optimized through the physical information neural network module, which is integrated into the adaptive envelope fitting module. Adaptive envelope fitting module: used to fit a benchmark loss envelope model based on a stable subset of multi-source data through quantile regression or deep quantization regression; Additional loss calculation module: used to calculate non-negative additional loss and its time accumulation in real time based on the baseline loss envelope model.

7. A computer device, characterized in that: include: One or more processors; A processor is used to store one or more programs; When one or more programs are executed by one or more processors, the method for real-time estimation of transformer additional losses based on quantile envelope and dynamic characteristics as described in any one of claims 1-5 is implemented.

8. A computer-readable storage medium, characterized in that, It contains a computer program, which, when executed, implements the method for real-time estimation of transformer additional losses based on quantile envelope and dynamic characteristics as described in any one of claims 1-5.

Images