State change-oriented micro-grid load intelligent evaluation method

Abstract

The invention relates to the technical field of micro-grids, in particular to a state change-oriented micro-grid load intelligent evaluation method, which comprises the following steps of: acquiring a load power data sequence of a micro-grid when the load state of the micro-grid changes; performing load change fitting on the load power data sequence by using an S-type jump function to construct a nonlinear fitting residual factor for representing a load jump process; calculating a data mean value of the load power data sequence to obtain a load fluctuation value sequence, and performing time axis inversion on the load fluctuation value sequence to construct a time domain reflection symmetry factor; and forming a two-dimensional feature vector by the nonlinear fitting residual factor and the time domain reflection symmetry factor, performing clustering identification on the two-dimensional feature vector, determining an event type corresponding to the load state change of the microgrid, and realizing adaptive classification identification of the load state change event type through clustering. And the load change evaluation efficiency and precision of the micro-grid are improved.

Description

Technical Field

[0001] This invention relates to the field of microgrid technology, and in particular to a method for intelligent load assessment of microgrids oriented towards state changes. Background Technology

[0002] Microgrids, as systems integrating distributed power sources, energy storage devices, energy conversion devices, and related load monitoring and protection devices, are crucial for the development of smart grids when operating safely, economically, and efficiently. Short-term load forecasting is a prerequisite for optimized microgrid scheduling, while accurate assessment of load state changes is fundamental for responding to system disturbances, ensuring power quality, and formulating dynamic control strategies. Changes in load state pose challenges to the stable operation of microgrids, such as the start-up and shutdown of large equipment, fluctuations in renewable energy output, or the occurrence of fault events. Therefore, intelligent methods capable of accurately identifying and assessing the nature of load state changes have significant theoretical and engineering application value.

[0003] In existing technologies, various methods have been used for load status assessment or anomaly detection in microgrids. Among them, a representative approach proposes a load forecasting and assessment method based on the coefficient of variation. This method calculates the coefficient of variation within a short sliding window, which is the ratio of the standard deviation to the arithmetic mean of the load data within the sliding window. This method quantifies sudden changes or short-cycle fluctuations in load data using the coefficient of variation. When the coefficient of variation exceeds a preset threshold, the system considers a significant change in load status to have occurred. By introducing the statistical indicator of the coefficient of variation, this method can capture dynamic changes in load to a certain extent, providing a feasible technical approach for load status assessment.

[0004] However, the load state assessment method based on the coefficient of variation has an inherent flaw in practical applications: it cannot effectively distinguish whether the load state change is due to the overall shift of the system operating baseline or the deterioration of the internal dynamic characteristics of the system near the current baseline. This confusion and misjudgment of the nature of the load change will affect the accuracy of the assessment results and mislead the subsequent energy management system to make incorrect control and scheduling decisions. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a method for intelligent load assessment of microgrids oriented towards state changes, in order to solve the problem of large errors in load state assessment of microgrids based on the coefficient of variation.

[0006] This invention provides a method for intelligent load assessment of microgrids oriented towards state changes, the method comprising the following steps: When the load state of the microgrid changes, the load power data sequence of the microgrid is acquired; The load power data sequence is fitted with an S-shaped jump function to construct a nonlinear fitting residual factor to characterize the load jump process; the mean of the load power data sequence is calculated, and the mean is used to zero-mean processing of each data point in the load power data sequence to obtain a load fluctuation value sequence composed of the load fluctuation values ​​of each data point; the load fluctuation value sequence is inverted on the time axis to construct a time-domain reflection symmetry factor. The nonlinear fitting residual factor and the time-domain reflection symmetry factor are combined to form a two-dimensional feature vector. The two-dimensional feature vector is clustered and identified using a preset event feature cluster center library to obtain the clustering result. Based on the clustering result, the event type corresponding to the load state change of the microgrid is determined.

[0007] Preferably, the step of fitting the load power data sequence with the S-shaped jump function to construct a nonlinear fitting residual factor for characterizing the load jump process includes: The load power data sequence is fitted with an S-shaped jump function to obtain a fitted value for each data point in the load power data sequence. Based on the root mean square error between each data point and its fitted value, a nonlinear fitting residual factor characterizing the load jump process is obtained. The calculation expression for the S-shaped jump function is as follows:

[0008] in, This represents the fitted value corresponding to the j-th data point in the load power data sequence. The data sequence number index represents the load power data sequence. This represents the first data point in the load power data sequence. This represents the predicted jump magnitude of the standard jump function. , This represents the last data point in the load power data sequence, where e represents the natural constant. This represents the predicted transition center time of the standard transition function. , The sampling interval is represented by N, the number of data points in the load power data sequence is represented by N, and 1 represents a constant. This indicates the steepness of the predicted curve of the standard jump function.

[0009] Preferably, the step of obtaining the nonlinear fitting residual factor characterizing the load jump process based on the root mean square error between each data point in the load power data sequence and its fitted value includes: Based on the fitted values ​​of each data point in the load power data sequence, the root mean square error is calculated, and the absolute value of the difference between the last data point and the first data point in the load power data sequence is obtained. Using the absolute value of the difference as the denominator and the root mean square error as the numerator, a nonlinear fitting residual factor for characterizing the load jump process is obtained.

[0010] Preferably, the step of using the data mean to perform zero-mean processing on each data point in the load power data sequence to obtain a load fluctuation value sequence composed of the load fluctuation values ​​of each data point includes: For any data point in the load power data sequence, the difference between the data point and the mean of the data is taken as the load fluctuation value of the data point. The load fluctuation value of each data point in the load power data sequence is obtained to form a load fluctuation value sequence.

[0011] Preferably, the step of reversing the load fluctuation value sequence on the time axis to construct a time-domain reflection symmetry factor includes: The load fluctuation value sequence is reversed to obtain a load fluctuation value inverted sequence. The products of two data points at the same position in the load fluctuation value sequence and the load fluctuation value inverted sequence are accumulated to obtain a first accumulated value. The squares of each data point in the load fluctuation value sequence are accumulated to obtain a second accumulated value. The time-domain reflectance symmetry factor is obtained by using the square of the first accumulated value as the numerator and the square of the second accumulated value as the denominator.

[0012] Preferably, the step of using a preset event feature cluster center library to cluster and identify the two-dimensional feature vector to obtain clustering results includes: Each cluster center in the event feature cluster center library corresponds to a known load state change event type. The Euclidean distance between the two-dimensional feature vector and the two-dimensional feature vector sample corresponding to each cluster center is calculated to obtain the minimum Euclidean distance. The clustering result of the two-dimensional feature vector is determined as the cluster center corresponding to the minimum Euclidean distance.

[0013] The beneficial effects of the embodiments of the present invention compared with the prior art are as follows: This invention acquires a load power data sequence of a microgrid when a load state change occurs. It then uses an S-shaped jump function to fit the load power data sequence to the load change, constructing a nonlinear fitting residual factor to characterize the load jump process. The invention calculates the data mean of the load power data sequence and uses this mean to zero-mean process each data point in the sequence, obtaining a load fluctuation value sequence composed of the load fluctuation values ​​of each data point. The load fluctuation value sequence is then inverted on the time axis to construct a time-domain reflection symmetry factor. The nonlinear fitting residual factor and the time-domain reflection symmetry factor are combined to form a two-dimensional feature vector. Using a pre-defined event feature cluster center library, the two-dimensional feature vector is clustered to obtain clustering results. Based on these clustering results, the event type corresponding to the load state change in the microgrid is determined. Specifically, a nonlinear fitting residual factor is constructed to characterize the load change process of the load power data sequence, and a time-domain reflection symmetry factor is constructed to characterize the load fluctuation law of the load power data sequence, so as to reflect the internal dynamic characteristics of the microgrid after entering a new stable operating state. Then, a two-dimensional feature vector composed of the nonlinear fitting residual factor and the time-domain reflection symmetry factor is used to characterize the comprehensive characteristics of load state change. Thus, adaptive classification and identification of load state change events are achieved through clustering, which improves the efficiency and accuracy of load change assessment of the microgrid. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of a method for intelligent load assessment of microgrids oriented towards state changes, provided in Embodiment 1 of the present invention. Figure 2 This is a schematic diagram illustrating the correlation between a load fluctuation value sequence and a load fluctuation value inversion sequence under a random noise scenario, provided by an embodiment of the present invention. Figure 3 This is a schematic diagram illustrating the relationship between a load fluctuation value sequence and a load fluctuation value reversal sequence under a periodic load scenario, as provided in an embodiment of the present invention. Detailed Implementation

[0016] Embodiments of this disclosure are described in detail below, with examples of these embodiments illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting it.

[0017] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure.

[0018] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0019] See Figure 1 This is a flowchart of a method for intelligent load assessment of microgrids oriented towards state changes, provided in Embodiment 1 of the present invention. Figure 1 As shown, the method may include: Step S101: When the load state of the microgrid changes, the load power data sequence of the microgrid is obtained.

[0020] Power quality monitoring devices deployed at the microgrid's grid connection point or key feeders continuously collect total active power data of the microgrid at a preset sampling frequency (10 times per second in this embodiment) to form a raw load power data sequence. Considering that the collected raw load power data sequence may contain missing values ​​due to communication interruptions or sensor failures, as well as outliers that significantly exceed the physical operating range, preprocessing of the raw load power data sequence is necessary to ensure the accuracy of subsequent evaluations. This preprocessing includes, but is not limited to, using linear interpolation to complete missing values ​​and removing outliers that exceed reasonable thresholds, thereby obtaining a continuous and complete time series sequence.

[0021] After obtaining the preprocessed time series, the absolute value of the difference between load power data at adjacent sampling times is calculated and recorded as the load change value. This load change value is used to monitor the load state changes of the microgrid in real time. Specifically, when the load change value is greater than or equal to 5% of the microgrid's rated power, it is determined that the microgrid may have experienced a load state change. Subsequently, a load power data sequence of fixed length N is obtained (i.e., the load power data sequence contains N data points, preferably set to N=60), for in-depth evaluation of the microgrid's load change state. It is worth noting that the load power data sequence is also a preprocessed time series.

[0022] Step S102: Use the S-shaped jump function to fit the load power data sequence to the load change, so as to construct a nonlinear fitting residual factor to characterize the load jump process; calculate the data mean of the load power data sequence, and use the data mean to perform zero mean processing on each data in the load power data sequence to obtain a load fluctuation value sequence composed of the load fluctuation value of each data; reverse the load fluctuation value sequence on the time axis to construct a time-domain reflection symmetry factor.

[0023] In existing technologies, load change status assessment of microgrids is typically based on the coefficient of variation (COP). However, when a microgrid experiences a high-amplitude steady-state jump in load, such as the startup of a high-power central air conditioning unit, causing the total load to steadily jump from a low-power platform to a high-power platform, the COP only briefly increases during the transient process of the jump. Once the load enters the new high-power stable platform, the numerator (standard deviation) in the COP calculation formula becomes extremely small, while the denominator (mean) becomes extremely large, causing the COP to plummet to an extremely low value. This can incorrectly assess a major load input event that has a significant impact on the microgrid as a stable state without abrupt changes. Conversely, when a microgrid experiences a highly volatile dynamic oscillation, such as multiple spot welding robots on an automated production line entering a periodic working mode, causing the total load mean to change little but exhibit dense high-frequency pulses, the COP will continuously output an extremely high value because the numerator in the calculation formula remains high while the denominator remains essentially unchanged. This can incorrectly assess a local disturbance event with a small increase in average power as a continuous high-amplitude abrupt change. Therefore, when a significant load change is detected, the load change status assessment method for microgrids based on the coefficient of variation cannot accurately distinguish whether the change is due to a single systematic load switching or a series of high-frequency dynamic disturbances. It cannot delve into the morphological level of the change process, leading to confusion and misjudgment of the load change status. To address this problem, this embodiment of the invention constructs a nonlinear fitting residual factor to effectively distinguish the process morphology of load changes.

[0024] The startup process of large inductive loads in a microgrid (such as chillers in a central air conditioning system, large industrial water pumps, or fans) is not an ideal instantaneous step. Affected by the electromagnetic transient effects of the motor windings and the mechanical inertia of the load itself, the active power rise curve exhibits a smooth, monotonically increasing transition process. The shape of this rise curve generally follows an S-shaped pattern, that is, the active power rises slowly in the initial stage of startup, the active power rises fastest in the middle stage, and the active power gradually flattens out in the final stage, eventually stabilizing at the new rated power level. In stark contrast, load fluctuations caused by periodic pulse loads or random disturbances exhibit drastic, irregular jumps in time-series data. Therefore, in this embodiment of the invention, a standard S-curve transition function is used to fit the load power data sequence to the load change. If the load change process corresponding to the load power data sequence is triggered by a smooth start-up of a large load, the load power data sequence will match the curve corresponding to the S-curve transition function, and the fitting residual will be relatively small. Conversely, if the load change process corresponding to the load power data sequence is triggered by chaotic high-frequency disturbances, fitting with the S-curve transition function will result in a bunch of oscillating data points, leading to poor fitting and a large fitting residual. Thus, by quantifying the magnitude of the fitting residual to distinguish the nature of the load change process, it is possible to effectively differentiate between systematic steady-state jump processes and dynamic high-frequency oscillation processes.

[0025] Specifically, firstly, the load power data sequence is fitted with a S-shaped transition function to obtain the fitted value for each data point in the load power data sequence. The calculation expression for the S-shaped transition function is as follows:

[0026] in, This represents the fitted value corresponding to the j-th data point in the load power data sequence. The data sequence number index represents the load power data sequence. This represents the first data point in the load power data sequence. This represents the predicted jump magnitude of the standard jump function. , This represents the last data point in the load power data sequence, where e represents the natural constant. This represents the predicted transition center time of the standard transition function. , The sampling interval is 0.1 seconds in this embodiment, N represents the number of data points in the load power data sequence, and 1 represents a constant. This represents the steepness of the predicted curve of the standard jump function, and is preferably set to 2.

[0027] Then, based on the root mean square error between each data point in the load power data sequence and its fitted value, a nonlinear fitting residual factor for characterizing the load jump process is obtained: The root mean square error is calculated based on the fitted value of each data point in the load power data sequence; the absolute value of the difference between the last data point and the first data point in the load power data sequence is obtained; using the absolute value of the difference as the denominator and the root mean square error as the numerator, the nonlinear fitting residual factor for characterizing the load jump process is obtained. The formula for calculating the nonlinear fitting residual factor is as follows:

[0028] in, represents the nonlinear fitting residual factor, N represents the number of data points in the load power data sequence, and | represents the absolute value sign.

[0029] It should be noted that, The root mean square error (RMSE) represents the difference between the load power data sequence and the fitted value obtained by fitting using the S-curve function, also known as the fitting residual. This residual is collected when the load change occurs during the startup of a large load. Will be with fitted values The data shows a high degree of agreement, with a very small difference between the two, resulting in a small fitting residual. This indicates that the load change process of the load power data sequence highly conforms to the smooth start-up of a large load, representing a normal steady-state jump. Conversely, when the load change is caused by irregular high-frequency disturbances, the load power data sequence will exhibit a back-and-forth jumping pattern. In this case, fitting with an S-shaped jump function will inevitably yield a huge fitting residual, which means... The value is large, which in turn increases the root mean square error (fitting residual), indicating that the load change process of the load power data sequence is highly consistent with the irregular high-frequency disturbance, which belongs to a dynamic disturbance.

[0030] Used for normalization, it measures the fit residual of the load power data series by using the actual change magnitude of the load power data series as a benchmark. This makes its value no longer affected by the absolute magnitude of the load jump, but rather by the goodness of fit. For example, a smooth start from 100kW to 200kW and a smooth start from 1000kW to 2000kW may have different fit residuals, but after normalization, the denominator (…) After normalization, the result is The values ​​will be very small; while an oscillation around 100kW ±50kW, although its fitting residual is smaller than that of a large smooth start, has a much smaller denominator ( The value is very small, after normalization The value will be relatively large. By decoupling the assessment of the load change process morphology from the assessment of the load change magnitude, the problem of characterization confusion caused by coupled calculation in existing technologies is fundamentally solved.

[0031] This yields a nonlinear fitting residual factor for characterizing the load jump process in the load power data sequence. This residual factor effectively distinguishes the morphology of the load change process, determining whether the load change in the load power data sequence is a systematic steady-state jump or a dynamic oscillation disturbance. However, to achieve a comprehensive assessment of the load change state, it is necessary to define the nature of the new steady state formed after the load change is complete. Specifically, after confirming that the microgrid has entered a new stable operating state, how can we distinguish whether the load fluctuations in this operating state originate from random background noise or are dynamic oscillation disturbances caused by equipment with inherent rhythmicity or periodicity? Accurate assessment of the nature of the new steady state is a crucial prerequisite for subsequent refined control and fault diagnosis.

[0032] Random background noise is temporally disordered and lacks a predictable structure; reversing its time series reveals no clear correlation with the original sequence. Conversely, dynamic disturbances caused by periodic loads such as variable frequency drive equipment exhibit periodic fluctuations and structural regularities in the time dimension, demonstrating temporal self-similarity. This means that reversing the original sequence along the time axis will result in a highly similar sequence in shape to the original. Therefore, this embodiment of the invention uses a time-domain reflection symmetry factor to characterize the load fluctuation patterns of the load power data sequence.

[0033] Specifically, firstly, the load fluctuation component of the load power data sequence is extracted: the mean of the load power data sequence is calculated, and for any data point in the load power data sequence, the difference between that data point and the mean is taken as the load fluctuation value of that data point. This process is repeated to obtain the load fluctuation value for each data point in the load power data sequence, forming a load fluctuation value sequence. Then, the sequence position of the load fluctuation value sequence is reversed to obtain a reversed load fluctuation value sequence. For example, if the load fluctuation value sequence is 123456, then the reversed load fluctuation value sequence is 654321. Next, the products of two data points at the same position in the load fluctuation value sequence and the reversed load fluctuation value sequence are accumulated to obtain a first accumulated value. The squares of each data point in the load fluctuation value sequence are accumulated to obtain a second accumulated value. The time-domain reflectance symmetry factor is obtained by using the square of the first accumulated value as the numerator and the square of the second accumulated value as the denominator.

[0034] The formula for calculating the time-domain reflectance symmetry factor is as follows:

[0035] in, Represents the time-domain reflectance symmetry factor. This represents the j-th load fluctuation value in the load fluctuation value sequence. This represents the j-th load fluctuation value in the load fluctuation value reversal sequence, and N represents the number of data points in the load fluctuation value sequence.

[0036] It should be noted that, The calculation is the square of the dot product between the load fluctuation value sequence and its own time-reversed sequence (load fluctuation value reversal sequence). When the load fluctuation in the new stable operating state is dominated by periodic or quasi-periodic equipment, the load fluctuation has an inherent symmetrical structure in time. At this time, the load fluctuation value sequence and its reversal sequence (load fluctuation value reversal sequence) will be highly similar in shape, resulting in a dot product value... This results in a very large absolute value, which, when squared, further amplifies this similarity. Conversely, when load fluctuations in a new stable operating state are irregular background noise, the load power data sequence lacks structural symmetry in time series, and the load fluctuation value sequence is statistically uncorrelated with its own time-reversed sequence (load fluctuation value reversal sequence), resulting in a dot product value. It will approach zero due to the random cancellation of positive and negative values, resulting in The value decreases.

[0037] The calculation uses the square of the load fluctuation value sequence as a normalization benchmark, eliminating the influence of the absolute amplitude of load fluctuation on the normalization process. The impact of value calculation. The value is essentially the fourth power of the correlation coefficient between the load fluctuation value sequence and the load fluctuation value inversion sequence, and its value is constrained to be between 0 and 1. The value is close to This indicates that the internal fluctuations of the new stable operating state exhibit a highly periodic or symmetrical structure, corresponding to the continuous operating state of the surrounding load; while a The value is close to This indicates that the internal fluctuations of the new stable operating state are random, corresponding to the state after the random background noise stabilizes.

[0038] For example: refer to Figure 2 This is a schematic diagram illustrating the correlation between the load fluctuation value sequence and the load fluctuation value inversion sequence under random noise conditions. Figure 2 The solid line represents the load fluctuation value sequence, and the dashed line represents the load fluctuation value reversal sequence. The calculation at this time... The value is 0.01242; conversely, refer to... Figure 3 This is a schematic diagram illustrating the correlation between the load fluctuation value sequence and the load fluctuation value inversion sequence under a periodic load scenario. Figure 3The solid line represents the load fluctuation value sequence, and the dashed line represents the load fluctuation value reversal sequence. The calculation at this time... The value is 0.90451.

[0039] Thus, we obtained the nonlinear fitting residual factor characterizing the load change process pattern of the load power data sequence and the time-domain reflection symmetry factor characterizing the internal dynamic characteristics of the new stable operating state.

[0040] Step S103: The nonlinear fitting residual factor and the time-domain reflection symmetry factor are combined to form a two-dimensional feature vector. Using the preset event feature cluster center library, the two-dimensional feature vector is clustered and identified to obtain the clustering results. Based on the clustering results, the event type corresponding to the load state change of the microgrid is determined.

[0041] In this embodiment of the invention, the nonlinear fitting residual factor and the time-domain reflection symmetry factor are combined to form a two-dimensional feature vector. This is used to characterize the comprehensive features of current load state changes. A pre-stored event feature cluster center library is obtained. Each cluster center in the library corresponds to a known load state change event type, such as: large load access events, large load disconnection events, pulse disturbance load access events, and line short-circuit triggered circuit breaker protection events, etc. The event feature cluster center library is constructed as follows: during microgrid operation, the nonlinear fitting residual factor and time-domain reflection symmetry factor are obtained for multiple load state changes and used to form two-dimensional feature vector samples, thus obtaining a sample set. Then, an unsupervised clustering algorithm (K-means clustering algorithm) is used to cluster the two-dimensional feature vector samples in the sample set. Since... and K-means clustering can effectively distinguish between load state change events of different natures. Therefore, the two-dimensional feature vector samples in the sample set exhibit high intra-class aggregation and inter-class segregation in the two-dimensional feature space. This allows the K-means clustering algorithm to identify clusters representing various typical events and store the cluster center C of each cluster in the event feature cluster center library. It is worth noting that the K-means clustering algorithm is existing technology and will not be described in detail here.

[0042] Then, using the event feature cluster center library, the two-dimensional feature vectors are clustered and identified to obtain the clustering results: the Euclidean distance between the two-dimensional feature vector V and the two-dimensional feature vector sample corresponding to each cluster center C is calculated to obtain the minimum Euclidean distance. The clustering result of the two-dimensional feature vector is determined as the cluster center corresponding to the minimum Euclidean distance, which is denoted as the target cluster center. Then, the event type corresponding to the load state change of the microgrid is determined as the event type corresponding to the target cluster center.

[0043] It should be noted that after determining the event type corresponding to the load state change in the microgrid, a structured assessment report containing detailed information about the event can be generated. This structured assessment report includes the event's occurrence time and the identified event type, and further includes a confidence score characterizing the classification reliability. This confidence score is calculated based on the Euclidean distance between the two-dimensional feature vector V and its target cluster center; the closer the Euclidean distance, the higher the confidence score. The generated structured assessment report is then sent to the microgrid's energy management system. Based on the event type and confidence score specified in the structured assessment report, the energy management system executes a pre-defined dynamic hierarchical response logic. For an event identified with high confidence as a high-amplitude steady-state jump, the energy management system automatically executes the matching Level 1 response plan, adjusting the energy storage output power. For an event with a low confidence score or that cannot be clearly attributed to any known cluster center, the energy management system marks it as an unknown or abnormal event and triggers a Level 2 or 3 response plan requiring manual intervention, issuing a high-level alarm and saving detailed field data for subsequent analysis.

[0044] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for intelligent load assessment of microgrids oriented towards state changes, characterized in that, The method includes: When the load state of the microgrid changes, the load power data sequence of the microgrid is acquired; The load power data sequence is fitted with an S-shaped jump function to construct a nonlinear fitting residual factor to characterize the load jump process; the mean of the load power data sequence is calculated, and the mean is used to zero-mean processing of each data point in the load power data sequence to obtain a load fluctuation value sequence composed of the load fluctuation values ​​of each data point; the load fluctuation value sequence is inverted on the time axis to construct a time-domain reflection symmetry factor. The nonlinear fitting residual factor and the time-domain reflection symmetry factor are combined to form a two-dimensional feature vector. The two-dimensional feature vector is clustered and identified using a preset event feature cluster center library to obtain the clustering result. Based on the clustering result, the event type corresponding to the load state change of the microgrid is determined.

2. The intelligent load assessment method for microgrids oriented towards state changes according to claim 1, characterized in that, The step of fitting the load power data sequence to load changes using an S-shaped jump function to construct a nonlinear fitting residual factor characterizing the load jump process includes: The load power data sequence is fitted with an S-shaped jump function to obtain a fitted value for each data point in the load power data sequence. Based on the root mean square error between each data point and its fitted value, a nonlinear fitting residual factor characterizing the load jump process is obtained. The calculation expression for the S-shaped jump function is as follows:

3. Among them, This represents the fitted value corresponding to the j-th data point in the load power data sequence. The data sequence number index represents the load power data sequence. This represents the first data point in the load power data sequence. This represents the predicted jump magnitude of the standard jump function. , This represents the last data point in the load power data sequence, where e represents the natural constant. This represents the predicted transition center time of the standard transition function. , The sampling interval is represented by N, the number of data points in the load power data sequence is represented by N, and 1 represents a constant. This indicates the steepness of the predicted curve of the standard jump function.

4. The intelligent load assessment method for microgrids oriented towards state changes according to claim 2, characterized in that, The step of obtaining the nonlinear fitting residual factor characterizing the load jump process based on the root mean square error between each data point in the load power data sequence and its fitted value includes: Based on the fitted values ​​of each data point in the load power data sequence, the root mean square error is calculated, and the absolute value of the difference between the last data point and the first data point in the load power data sequence is obtained. Using the absolute value of the difference as the denominator and the root mean square error as the numerator, a nonlinear fitting residual factor for characterizing the load jump process is obtained.

5. The intelligent load assessment method for microgrids oriented towards state changes according to claim 1, characterized in that, The step of using the data mean to perform zero-mean processing on each data point in the load power data sequence to obtain a load fluctuation value sequence composed of the load fluctuation values ​​of each data point includes: For any data point in the load power data sequence, the difference between the data point and the mean of the data is taken as the load fluctuation value of the data point. The load fluctuation value of each data point in the load power data sequence is obtained to form a load fluctuation value sequence.

6. The intelligent load assessment method for microgrids oriented towards state changes according to claim 1, characterized in that, The step of reversing the load fluctuation value sequence on the time axis to construct a time-domain reflection symmetry factor includes: The load fluctuation value sequence is reversed to obtain a load fluctuation value inverted sequence. The products of two data points at the same position in the load fluctuation value sequence and the load fluctuation value inverted sequence are accumulated to obtain a first accumulated value. The squares of each data point in the load fluctuation value sequence are accumulated to obtain a second accumulated value. The time-domain reflectance symmetry factor is obtained by using the square of the first accumulated value as the numerator and the square of the second accumulated value as the denominator.

7. The intelligent load assessment method for microgrids oriented towards state changes according to claim 1, characterized in that, The method of using a preset event feature cluster center library to perform clustering and identification on the two-dimensional feature vector to obtain clustering results includes: Each cluster center in the event feature cluster center library corresponds to a known load state change event type. The Euclidean distance between the two-dimensional feature vector and the two-dimensional feature vector sample corresponding to each cluster center is calculated to obtain the minimum Euclidean distance. The clustering result of the two-dimensional feature vector is determined as the cluster center corresponding to the minimum Euclidean distance.

Images