Electric power project data fusion method and system

Abstract

The invention relates to the technical field of data processing, in particular to an electric power project data fusion method and system. The method comprises the following steps: collecting and preprocessing multi-dimensional time sequence data of a distributed power supply and an energy storage system; extracting features of the preprocessed data; an adaptive forgetting factor is introduced to carry out adaptive Kalman filtering processing on the characteristic data, the adaptive forgetting factor combines a current error, an error change trend and a historical error weighted sum difference value, and a prediction covariance matrix and a measurement noise covariance matrix are dynamically adjusted, so that the response capability of filtering to system mutation is improved; and fusing the optimal estimation state of each feature by adopting a D-S evidence theory so as to accurately judge the operation state of the electric power project. According to the method, the problem that estimation is lagged when the state is suddenly changed in traditional Kalman filtering is effectively solved, the accuracy and the reliability of judging the operation state of the electric power project are improved, and an effective guarantee is provided for stable operation of a micro-grid system.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for data fusion in power projects. Background Technology

[0002] As a small-scale power generation and distribution system that integrates multiple renewable energy sources, microgrids are moving from demonstration projects to large-scale applications driven by the current growth in power project investment and dual carbon targets. Microgrid systems are typically composed of distributed power sources, such as photovoltaic and wind power, as well as various equipment such as energy storage devices and energy conversion devices. The wide variety of equipment and diverse operating modes in microgrid systems lead to prominent multi-source coordination and control issues.

[0003] In the monitoring and diagnostic process of microgrids, data fusion is a crucial step. Data from different sensors and devices needs to be effectively integrated to accurately assess the system state. Currently, Kalman filtering algorithms are typically used to process the collected data before data fusion, combining historical predictions with current measurements to obtain the optimal state estimate. However, traditional Kalman filtering algorithms are based on the assumption that the system state changes steadily and the noise characteristics are fixed.

[0004] In the actual operating environment of microgrids, the system state often undergoes sudden changes. For example, the light intensity may drop abruptly due to cloud cover, or a large inductive load may be suddenly connected, which can cause drastic fluctuations in photovoltaic output power or system voltage. Therefore, the traditional Kalman filter algorithm cannot adjust the trust weights of historical prediction data and current measured data in a timely manner. It will still highly trust the predictions made based on historical data in a stable state, causing the filtering results to deviate from the true state and produce estimation lag. The lagging and distorted data will directly affect the quality of subsequent data fusion, thereby causing incorrect fault diagnosis or inappropriate control strategies, which in turn endangers the stability and reliability of the entire microgrid system. Summary of the Invention

[0005] To address the problem that traditional Kalman filtering suffers from estimation lag when processing power project data with abrupt changes, leading to reduced quality of fused data, this invention provides a power project data fusion method and system.

[0006] In a first aspect, the present invention provides a method for data fusion in power projects, employing the following technical solution: A data fusion method for power projects includes: collecting multi-dimensional time-series data of distributed power sources and energy storage systems; preprocessing the multi-dimensional time-series data to obtain preprocessed data, extracting features from the preprocessed data to obtain feature data; processing the feature data using Kalman filtering to obtain the optimal estimated state of each feature at the corresponding time; the Kalman filtering includes an adaptive forgetting factor, where the difference between the measured value and the predicted value at each time is recorded as the error, and the adaptive forgetting factor is determined based on the error at the current time, the trend of the error at the current time, and the difference between the weighted sum of the errors at the current time and the previous time within a set window; using the adaptive forgetting factor to adjust the prediction covariance matrix and the measurement noise covariance matrix, wherein the adjusted prediction covariance matrix is ​​negatively correlated with the adaptive forgetting factor, and the adjusted measurement noise covariance matrix is ​​positively correlated with the product of the initial measurement noise covariance matrix and the exponential term containing the adaptive forgetting factor; and fusing the optimal estimated state using DS evidence theory to determine the operating status of the power project.

[0007] The beneficial effects are as follows: By introducing an adaptive forgetting factor, the Kalman filter parameters are adaptively adjusted. Based on a comprehensive analysis of the current error, error change trend, and weighted sum of historical errors, the trust weights of the filtering algorithm on historical data and current measurements are dynamically adjusted. When the system state undergoes a sudden change, the error increases and the change trend becomes obvious. The adaptive forgetting factor increases accordingly, reducing the weight of the prediction covariance matrix and increasing the weight of the measurement noise covariance matrix. This makes the filter rely more on current measurement data, effectively reducing estimation lag. When the system state is stable, the adaptive forgetting factor decreases, maintaining appropriate trust in historical predictions and ensuring filter stability. By adjusting the prediction covariance matrix and measurement noise covariance matrix through the adaptive forgetting factor, precise control of the filtering parameters is achieved, improving the adaptability of the Kalman filter under sudden data changes in power projects and effectively solving the estimation lag problem. Combining DS evidence theory with multi-source data fusion fully utilizes the optimal estimated state information of each feature, improving the accuracy and reliability of power project operation state judgment and providing effective technical support for the stable operation of microgrid systems.

[0008] Furthermore, the acquisition of multi-dimensional time-series data of distributed power sources and energy storage systems includes: acquiring current and voltage data using current transformers and voltage transformers; acquiring irradiance data of photovoltaic systems using irradiance sensors; acquiring wind speed data of wind power systems using wind speed sensors; acquiring battery temperature data of energy storage systems using temperature sensors; and calculating the state of charge data of energy storage systems using the ampere-hour integration method.

[0009] Furthermore, the acquisition of multi-dimensional time-series data of distributed power sources and energy storage systems includes: acquiring current and voltage data using current transformers and voltage transformers; acquiring irradiance data of photovoltaic systems using irradiance sensors; acquiring wind speed data of wind power systems using wind speed sensors; acquiring battery temperature data of energy storage systems using temperature sensors; and calculating the state of charge data of energy storage systems using the ampere-hour integration method.

[0010] Furthermore, features of the preprocessed data are extracted to obtain feature data, including: extracting the active power change rate, reactive power change rate, light intensity change trend, and wind speed change trend of the distributed power source; and extracting the state of charge change rate, charge / discharge current change rate, and battery temperature change trend of the energy storage system.

[0011] Furthermore, the adaptive forgetting factor satisfies: In the formula, For a moment The adaptive forgetting factor For a moment The error, For a moment The difference between the error at the previous moment and the error at the previous moment. For a moment The difference between the weighted sum of errors from multiple historical moments within the set window and the previous moment. This is a moderating coefficient used to adjust the influence of the trend of error on the adaptive forgetting factor. This is the sensitivity coefficient used to control the degree to which the adaptive forgetting factor is sensitive to error. The error threshold used to adjust the variation range of the adaptive forgetting factor, For hyperparameters, For symbolic functions, It is a natural exponential function.

[0012] The beneficial effects are as follows: by constructing a composite function containing an error trend term and an exponential decay term, the evaluation of the adaptive forgetting factor is realized. The error trend term can effectively reflect the direction and magnitude of error changes, while the exponential decay term ensures that the forgetting factor changes within a reasonable range. The sign function reflects the relationship between the current error change and the historical error trend. When the current error change is consistent with the historical trend, the forgetting factor is adjusted accordingly. When abnormal changes occur, the parameters can be responded to and adjusted in a timely manner.

[0013] Furthermore, the symbol function satisfies: in response to Greater than 0, The value is 1, otherwise, The value of is 0.

[0014] Furthermore, the weighted sum of the errors satisfies: In the formula, For a moment The weighted sum of errors from multiple historical moments within a set window. For indexing time, For a moment Before The first historical moment At that moment, To set the number of historical moments within the window, For a moment Before The first historical moment Error at any given moment This is the forgetting rate coefficient. It is a natural exponential function.

[0015] The beneficial effects are as follows: by constructing a weighted sum model with exponential decay weights, the fusion of historical error information is achieved. The exponential decay term ensures that the more recent the error, the greater its impact on the current state, and the more distant the error, the smaller its impact. The forgetting rate coefficient controls the rate of decay, making the use of historical information more reasonable. It effectively captures the dynamic trend of error changes, providing a reliable historical reference for the calculation of the adaptive forgetting factor. Through the weighted summation, the random influence that a single historical point error may bring is avoided, improving the stability and accuracy of error trend analysis.

[0016] Furthermore, the adjusted prediction covariance matrix satisfies: In the formula, For a moment The adjusted prediction covariance matrix, For a moment The adaptive forgetting factor For a moment Posterior covariance Through the state transition matrix Transmitted to time Prior uncertainty, For a moment The process noise covariance matrix.

[0017] The beneficial effects are as follows: by dividing the traditional prediction covariance matrix by the adaptive forgetting factor, the prediction uncertainty is dynamically adjusted. When the forgetting factor increases, the prediction covariance decreases, reducing the dependence on historical predictions and improving the response speed to changes in the current state.

[0018] Furthermore, the adjusted measurement noise covariance matrix satisfies: In the formula, For a moment The adjusted measurement noise covariance matrix, The initial measurement noise covariance matrix, This is the amplification factor used to control the maximum adjustment magnitude of the adjusted measurement noise covariance matrix. For a moment The adaptive forgetting factor This is a response factor used to achieve differentiated responses to the anomaly degree of the system through the properties of the exponential function.

[0019] The beneficial effects are as follows: by constructing a composite function containing the initial noise covariance and an adaptive adjustment term, the measurement noise covariance matrix is ​​adjusted. The exponential term enables a differentiated response to the system's anomaly. When the adaptive forgetting factor is small, the adjustment term increases, and the measurement noise covariance increases accordingly. When the system state changes abruptly, the adaptive forgetting factor decreases, the exponential term increases, and the adjusted measurement noise covariance matrix increases accordingly, reducing the confidence in the current measurement and avoiding excessive influence of abnormal measurements on the filtering results. When the system state is stable, the measurement noise covariance is close to the initial value, maintaining the filtering accuracy.

[0020] Secondly, this invention provides a power project data fusion system, which adopts the following technical solution: A power project data fusion system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned power project data fusion method is implemented.

[0021] By adopting the above technical solution, a computer program is generated from the above-mentioned power project data fusion method and stored in a memory so that it can be loaded and executed by a processor. Terminal equipment can then be made based on the memory and processor for convenient use.

[0022] The present invention has the following technical effects: (1) Breaking through the limitations of fixed trust weights in traditional Kalman filtering algorithms, the system achieves real-time adjustment of the trust in historical predictions and current measurements through an adaptive forgetting factor. The calculation of the forgetting factor integrates the current error, error change trend, and weighted sum and difference of historical errors, which can accurately capture sudden changes in the state of microgrids, such as sudden drop in light intensity or sudden load transfer. When a sudden change occurs in the system, the error increases and the trend becomes steep, and the forgetting factor will be adjusted accordingly. On the one hand, the adjusted prediction covariance matrix is ​​negatively correlated with the forgetting factor. At this time, the prediction covariance increases, reducing the trust in historical predictions based on the stationary assumption. On the other hand, the adjusted measurement noise covariance matrix is ​​positively correlated with the exponential term containing the forgetting factor, strengthening the dependence on the current measurement that reflects the true state of the sudden change. This avoids the estimation lag caused by excessive trust in historical data in traditional Kalman filtering algorithms, enabling the filtering results to quickly fit the true state.

[0023] (2) The precise control of the covariance matrix by the adaptive forgetting factor enables the Kalman filter to output high-quality optimal estimates in both stationary and abrupt states. When stationary, the forgetting factor maintains a reasonable value, the predicted covariance is small, and the filtering result takes into account both historical trends and measurement accuracy, suppressing random noise. When abrupt, the covariance matrix is ​​quickly adjusted, and the current measurement is given priority to track state changes and avoid distortion. Compared with the lag-distorted data output by traditional filtering, the optimal estimated state obtained by this invention is closer to the actual operation of the microgrid, providing a highly reliable data source foundation for subsequent multi-source data fusion and reducing fusion errors.

[0024] (3) Combining the DS evidence theory to fuse multidimensional optimal estimated states, such as distributed power generation, energy storage SOC, system voltage, etc., fully leveraging its advantages in handling uncertainties of multi-source information, the optimal estimated states of different dimensions each carry local state information of the microgrid. The DS evidence theory can integrate the confidence of information of each dimension through evidence combination rules, effectively suppressing the random error or local interference of single-dimensional data, such as the instantaneous failure of a certain sensor, avoiding the fusion deviation caused by misjudgment of single data, making the judgment of the operating status of power projects more comprehensive and accurate, and reducing the probability of misjudgment. Attached Figure Description

[0025] Figure 1 This is a flowchart of a power project data fusion method according to an embodiment of the present invention. Detailed Implementation

[0026] 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, not all, of the embodiments of the present invention. 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.

[0027] This invention discloses a data fusion method for power projects, referring to... Figure 1 This includes steps S01-S04: S01: Collect multi-dimensional time-series data of distributed power sources and energy storage systems, and preprocess the multi-dimensional time-series data to obtain preprocessed data.

[0028] It is important to note that in order to comprehensively and accurately assess the operating status of a microgrid, it is essential to first collect real-time, continuous operational data from each key device node. However, sensors are inevitably affected by factors such as electromagnetic interference, environmental changes, or their own aging during the data acquisition process, resulting in noise, outliers, or data loss due to communication failures in the raw data. Without preprocessing, this flawed data will interfere with the accuracy of subsequent feature extraction and may mislead the state estimation of the Kalman filter, ultimately leading to distorted fusion results. Therefore, this step, through a systematic preprocessing procedure, aims to purify the data and provide a high-quality, complete, and reliable dataset for subsequent steps.

[0029] Specifically, data is collected from the core units of the microgrid: For distributed power sources, such as photovoltaic arrays and wind turbines, data such as current, voltage, illuminance (unit: lx) and wind speed (unit: m / s) are collected through corresponding current transformers, light intensity sensors and wind speed sensors; For energy storage systems, such as battery packs, voltage, current and temperature sensors, such as thermocouples, are used to collect the battery's terminal voltage, charging and discharging current and battery temperature (unit: °C), and the state of charge (SOC) is calculated using methods such as the ampere-hour integration method. The collected data from various dimensions constitute the original data matrix in time series; Preprocess the original data matrix: First, median filtering is used to process common impulse noise or isolated singularities in sensor data. Since median filtering uses the median value within the data window to replace the value of the center point, it is not easily affected by extreme values ​​and can effectively filter out noise while better protecting the edge information of the signal. Applying the 3σ principle in statistics to correct outliers involves calculating the mean and standard deviation of each data sequence, and then removing outliers that exceed the mean. Data points within a range of (multiple standard deviations) are identified as outliers and replaced with nearby normal data or the mean to eliminate erroneous data that deviate significantly from the normal fluctuation range due to momentary sensor failure or major interference. Linear interpolation is used to fill in missing data. For missing sequence data caused by communication interruption or other reasons, the value of the missing point is linearly estimated by using known data points at two time points before and after the missing point. This ensures the continuity of the data sequence and avoids the problem of not being able to calculate features such as the rate of change due to data breakpoints. After the above denoising, outlier correction and missing value imputation, the output is an optimized data matrix, as exemplified in Table 1, i.e., the preprocessed data.

[0030] Table 1

[0031] S02: Extract the features of the preprocessed data to obtain feature data.

[0032] It's important to note that directly using preprocessed data for state estimation presents two main problems: first, the data has high dimensionality, containing a large amount of redundant information and resulting in a heavy computational load; second, the dimensions and numerical ranges of different physical quantities vary significantly—for example, voltage is on the order of hundreds of volts, while the rate of change of current may be in the single digits. Without processing, features with larger numerical values ​​will dominate the calculation results, obscuring smaller but equally important features. Therefore, this step is necessary to reduce dimensionality and highlight key information through feature extraction, and to eliminate the influence of dimensions through normalization, making all features comparable and thus improving the efficiency and accuracy of subsequent filtering and fusion algorithms.

[0033] Specifically, feature extraction reveals that the trend of change is a better indicator of the system's dynamic characteristics than the original instantaneous values.

[0034] Therefore, the following key features are mainly extracted: For distributed power sources, extract the active power change rate, reactive power change rate, light intensity change trend, wind speed change trend, etc. For energy storage systems, extracting features such as the rate of change of state of charge (SOC), the rate of change of charge and discharge current, and the trend of battery temperature change reflects the working status and health of the energy storage unit. For example, the rate of change of current can be calculated by subtracting the current at the previous moment from the current at the current moment and then dividing by the time interval. The extracted feature data is normalized: This embodiment uses the min-max normalization method to linearly map all data of each feature to... Within the range; The above operations output a feature data matrix, as exemplified in Table 2, which contains the feature data. Each row represents a multidimensional feature vector at a given time point, and each column represents a normalized feature index, providing ideal input for subsequent Kalman filtering.

[0035] Table 2

[0036] S03: Obtain the optimal state estimate based on adaptive Kalman filtering.

[0037] It should be noted that while preprocessing can filter out static noise and isolated outliers, it cannot effectively handle dynamic noise and system mutations in microgrid operation. For example, errors inherent in the photovoltaic power model and cumulative biases in battery SOC estimations are time-dependent. Although traditional Kalman filtering can handle this type of dynamic noise, its core flaw lies in assuming a fixed system model and noise characteristics. When faced with real-world mutations such as sudden changes in illumination or load switching, its fixed model parameters lead to over-reliance on historical predictions, failing to adapt quickly to actual changes in conditions and resulting in estimation lag. The core objective of this step is to fundamentally improve the traditional Kalman filter, enabling it to adaptively adjust the confidence levels of both internal and external models, thereby outputting more accurate data that better reflects the real-world conditions in the complex and ever-changing microgrid environment. Therefore, by evaluating the magnitude of errors, the trend of error changes, and the accumulation of historical errors in the filtering process in real time, an adaptive forgetting factor is constructed as a dynamic adjustment knob. On the one hand, it acts on the prediction covariance matrix to adjust the confidence in the prediction of the system model; on the other hand, it acts on the measurement noise covariance matrix to adjust the confidence in the current sensor measurement. Through the dual-channel collaborative adjustment mechanism, the Kalman gain is indirectly optimized, so that it relies more on historical predictions to smooth noise when the system is stable, and relies more on current measurements to quickly track changes when the system changes abruptly. Ultimately, the optimal balance between historical prediction and real-time measurement is achieved, and the optimal estimated state value is output.

[0038] S301: Determine the adaptive forgetting factor.

[0039] The difference between the measured value and the predicted value at each moment is recorded as the error. Based on the error at the current moment, the trend of the error at the current moment, and the difference between the weighted sum of the errors at the current moment and the previous moment within the set window, the adaptive forgetting factor at the current moment is determined.

[0040] Specifically, the adaptive forgetting factor satisfies: ; In the formula, For a moment The adaptive forgetting factor For a moment The error, For a moment The difference between the error at the previous moment and the error at the current moment represents the change in error. This is used to determine whether the system is deteriorating (error is increasing) or improving (error is decreasing), distinguishing between a continuous abrupt change and transient noise. This indicates that the error is increasing and the system may experience a sudden change. The result is relatively small, so avoid over-amplification of the trend adjustment term and prevent noise interference. This indicates that the error is decreasing, the system is stabilizing, and it is operating normally. The results are relatively large, and even small errors can be detected, thus providing early warnings; For a moment The difference between the weighted sum of errors from multiple historical moments within the set window and the previous moment represents the historical error change term; For example, the moderating coefficient used to adjust the influence of the error trend on the adaptive forgetting factor is... ; For example, a sensitivity coefficient is used to control the sensitivity of the adaptive forgetting factor to error. ; For example, an error threshold is used to adjust the variation range of the adaptive forgetting factor. Three times the standard deviation of the measured value; For example, hyperparameters. Hyperparameters exist to prevent At this time, the formula becomes meaningless; For symbolic functions, responding to A value greater than 0 indicates that the error is gradually increasing. The value of is 1; otherwise, it indicates that the error is gradually decreasing or there is no error. The value of is 0; It is a natural exponential function.

[0041] The construction logic of the above relation is shown in Table 3: Table 3

[0042] Specifically, the weighted sum of the errors satisfies: ; In the formula, For a moment The weighted sum of errors from multiple historical moments within a set window. For indexing time, For a moment Before The first historical moment At that moment, To set the number of historical moments within a window, for example, if , ,but from Beginning, to End, meaning only the errors at times 7, 8, and 9 are calculated, because recent data better reflects the current state; For a moment Before The first historical moment Error at any given moment For example, the forgetting rate coefficient. This is used to control the decay rate of historical errors; the smaller the value, the faster the forgetting. It is a natural exponential function.

[0043] S302: Use the adaptive forgetting factor to adjust the prediction covariance matrix.

[0044] It should be noted that traditional Kalman filtering assumes the system model is accurate and the process noise is stable. Therefore, it treats all historical data equally, assuming that the contribution of historical data to the current prediction is uniform. However, in the dynamically changing environment of microgrids, the system model may undergo sudden changes, such as sudden load changes or sudden changes in distributed power output. If all historical data is still treated equally, it will lead to an inability to quickly track the actual state when the system undergoes sudden changes. Therefore, this step adjusts the prediction covariance matrix in the Kalman filter.

[0045] The adaptive forgetting factor is used to adjust the prediction covariance matrix, and the adjusted prediction covariance matrix is ​​negatively correlated with the adaptive forgetting factor.

[0046] Specifically, the adjusted prediction covariance matrix satisfies: ; In the formula, For a moment The adjusted prediction covariance matrix, For a moment The adaptive forgetting factor For a moment Posterior covariance Through the state transition matrix Transmitted to time Prior uncertainty, For a moment The process noise covariance matrix.

[0047] in, An increase (close to 1) indicates that the error is small and there is no sign of a sustained increase or maintenance of a high level; the system is stable or has only minor noise. As the Kalman gain decreases, the Kalman gain also decreases. According to the traditional Kalman gain calculation formula, the Kalman gain increases with... Monotonically increasing indicates that historical estimates are given greater weight in state updates, and there is more reliance on historical predictions; A decrease indicates that the error is increasing, and the system may experience a sudden change. An increase in Kalman gain indicates less trust in historical data and a greater weighting of current observations.

[0048] S303: Use the adaptive forgetting factor to adjust the measurement noise covariance matrix.

[0049] It should be noted that the measurement reliability of sensors decreases significantly under adverse weather conditions, near the end of their lifespan, or when subjected to electromagnetic interference. The measurement noise covariance in traditional Kalman filtering cannot reflect this change. When sensor readings are unreliable, the filter will still assign excessive weights to the Kalman gain, leading to deviations in the final output. Therefore, this step adjusts the measurement noise covariance matrix in the Kalman filter.

[0050] The adaptive forgetting factor is used to adjust the measurement noise covariance matrix, and the adjusted measurement noise covariance matrix is ​​positively correlated with the product of the initial measurement noise covariance matrix and the exponential term containing the adaptive forgetting factor.

[0051] Specifically, the adjusted measurement noise covariance matrix satisfies: ; In the formula, For a moment The adjusted measurement noise covariance matrix, The initial measurement noise covariance matrix represents the inherent noise level of the sensor under ideal and stable operating conditions, and is determined through the sensor technical manual. For example, an amplification factor is used to control the maximum adjustment magnitude of the adjusted measurement noise covariance matrix. ; For a moment The adaptive forgetting factor For example, a response factor used to differentiate responses to system anomalies through the properties of an exponential function, .

[0052] in, The item represents the system anomaly index, when tending to hour, A value approaching 0 indicates that the error is small and there is no sign of continued increase or maintenance of a high level; the system is stable, meaning it is close to the reference measurement noise. When it approaches 0, A value approaching 1 indicates a larger error and the possibility of sudden changes, which significantly increases noise estimation. This is relevant to two characteristics of microgrid anomaly scenarios: frequent small disturbances, such as normal load changes and slight fluctuations in illumination, and occasional large sudden changes, such as faults and the switching on / off of large equipment. To achieve different effects, frequent small disturbances are filtered out without overreacting, while occasional large abrupt changes require a rapid and accurate response to amplify their impact. Therefore, when... Decrease, and the error increases. As the Kalman gain increases, the system becomes more unstable, the measurement becomes less reliable, and the Kalman gain decreases. According to the traditional Kalman gain calculation formula, the Kalman gain increases with... Monotonically decreasing; when Increase, and the error decreases. Decrease the Kalman gain, and the Kalman gain increases.

[0053] S304: The feature data is processed using an adaptive Kalman filter to obtain the optimal estimated state of each feature at the corresponding time.

[0054] pass Let the prediction of uncertainty be used With the measurement of uncertainty The linkage was achieved, and then the adaptive Kalman filter was used to process the feature data to obtain the optimal estimated state of each feature at the corresponding time.

[0055] S04: Use DS evidence theory to fuse the optimal estimated state to determine the operating status of the power project.

[0056] It should be noted that in microgrids, the data from distributed power sources and energy storage systems often contain significant uncertainties due to sensor accuracy limitations, environmental interference, and the ambiguity of equipment operating status, such as measurement errors and state ambiguities. DS evidence theory is an effective mathematical tool for handling uncertain information. By defining basic probability assignment functions, trust functions, and likelihood functions, it fuses multi-source evidence—in this embodiment, the optimal estimated state value output by the Kalman filter—with other characteristic data such as temperature trends, thereby deriving a more reliable decision basis. Essentially, it transforms the uncertainty of multi-source data into a probabilistic judgment of the system state, rather than directly processing numerical values.

[0057] Specifically, a basic probability assignment function is first constructed for each data source, such as photovoltaic output sensor and battery SOC estimator. This function assigns a basic probability value to each possible proposition, such as normal system operation, voltage exceeding limit, and abnormal power fluctuation, reflecting the degree of support of the data source for the proposition. These probability assignments do not require the summation to be one as in traditional probability theory, allowing explicit modeling of the unknown and uncertainty. Subsequently, the Dempster synthesis rule is used to fuse the basic probability assignments of multiple data sources. By calculating the intersection and conflict degree between evidence, the support probability of each data source for the same proposition is comprehensively analyzed, effectively eliminating contradictions between evidence and deriving the fused probability distribution. For example, when the photovoltaic power sensor indicates an abnormal output while the battery temperature sensor indicates that the system is normal, the Dempster rule can quantify the conflict between the two and output the credibility of a comprehensive proposition. The fusion result output is an updated basic probability assignment function, which more accurately describes the probability of the microgrid being in different operating states, providing quantitative support for subsequent state judgments, such as fault diagnosis and operating mode switching. Through DS evidence theory, the system can integrate redundant and complementary information, improving the robustness of decision-making in uncertain environments. Finally, the fused data is applied to the operation optimization and fault diagnosis of the microgrid. Through the analysis and processing of the fused data, a basis for the formulation of microgrid control strategies is provided, system faults are detected in a timely manner, and corresponding measures are taken to ensure the stable and efficient operation of the microgrid, thus completing the data fusion of the power project.

[0058] This invention also discloses a power project data fusion system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a power project data fusion method according to the present invention.

[0059] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0060] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for data fusion in power projects, characterized in that, include: Collect multi-dimensional time-series data of distributed power sources and energy storage systems, and preprocess the multi-dimensional time-series data to obtain preprocessed data; Extract features from the preprocessed data to obtain feature data; The feature data is processed using Kalman filtering to obtain the optimal estimated state of each feature at the corresponding time. The Kalman filter includes an adaptive forgetting factor, which records the difference between the measured value and the predicted value at each time as the error. The adaptive forgetting factor is determined based on the error at the current time, the trend of the error at the current time, and the difference between the weighted sum of the errors at the current time and the previous time within a set window. The adaptive forgetting factor is used to adjust the prediction covariance matrix and the measurement noise covariance matrix. The adjusted prediction covariance matrix is ​​negatively correlated with the adaptive forgetting factor, and the adjusted measurement noise covariance matrix is ​​positively correlated with the product of the initial measurement noise covariance matrix and the exponential term containing the adaptive forgetting factor. The optimal estimated state is fused using the DS evidence theory to determine the operating status of the power project.

2. The power project data fusion method according to claim 1, characterized in that, The collection of multi-dimensional time-series data from distributed power sources and energy storage systems includes: Current and voltage data are collected using current transformers and voltage transformers; Use a light intensity sensor to collect light intensity data from the photovoltaic system; Use wind speed sensors to collect wind speed data from wind power systems; Use temperature sensors to collect battery temperature data from the energy storage system; The state of charge (SOC) data of the energy storage system is calculated using the ampere-hour integral method.

3. The power project data fusion method according to claim 1, characterized in that, The preprocessing includes: Median filtering is used to denoise the time series data in each dimension; use The principle is to correct outliers in the denoised data. Linear interpolation was used to fill in the missing values ​​in the corrected data.

4. The power project data fusion method according to claim 1, characterized in that, Extracting features from the preprocessed data to obtain feature data includes: Extract the active power change rate, reactive power change rate, light intensity change trend, and wind speed change trend of distributed power sources; Extract the rate of change of state of charge, the rate of change of charge and discharge current, and the trend of battery temperature change of the energy storage system.

5. The power project data fusion method according to claim 1, characterized in that, The adaptive forgetting factor satisfies: ； In the formula, For a moment The adaptive forgetting factor For a moment The error, For a moment The difference between the error at the previous moment and the error at the previous moment. For a moment The difference between the weighted sum of errors from multiple historical moments within the set window and the previous moment. This is a moderating coefficient used to adjust the influence of the trend of error on the adaptive forgetting factor. This is the sensitivity coefficient used to control the degree to which the adaptive forgetting factor is sensitive to error. The error threshold used to adjust the variation range of the adaptive forgetting factor, For hyperparameters, For symbolic functions, It is a natural exponential function.

6. The power project data fusion method according to claim 5, characterized in that, The symbol function satisfies: In response to Greater than 0, The value is 1, otherwise, The value of is 0.

7. A power project data fusion method according to claim 1 or 5, characterized in that, The weighted sum of the errors satisfies: ； In the formula, For a moment The weighted sum of errors from multiple historical moments within a set window. For indexing time, For a moment Before The first historical moment At that moment, To set the number of historical moments within the window, For a moment Before The first historical moment Error at any given moment This is the forgetting rate coefficient. It is a natural exponential function.

8. The power project data fusion method according to claim 1, characterized in that, The adjusted prediction covariance matrix satisfies: ； In the formula, For a moment The adjusted prediction covariance matrix, For a moment The adaptive forgetting factor For a moment Posterior covariance Through the state transition matrix Transmitted to time Prior uncertainty, For a moment The process noise covariance matrix.

9. The power project data fusion method according to claim 1, characterized in that, The adjusted measurement noise covariance matrix satisfies: ； In the formula, For a moment The adjusted measurement noise covariance matrix, The initial measurement noise covariance matrix, This is the amplification factor used to control the maximum adjustment magnitude of the adjusted measurement noise covariance matrix. For a moment The adaptive forgetting factor This is a response factor used to achieve differentiated responses to the anomaly degree of the system through the properties of the exponential function.

10. A power project data fusion system, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a power project data fusion method according to any one of claims 1-9.

Images