New energy power generation state monitoring method and system

Abstract

The invention discloses a new energy power generation state monitoring method and system, and the method comprises the steps: collecting first object comprehensive data, and carrying out the first preprocessing; and constructing a first task graph according to the pre-processed data, and extracting a vector. And dynamically adjusting the vector, constructing a calculation model, and optimizing the first object power generation data. And constructing a visual interface to display the first object power generation data, and storing new first object comprehensive data generated by collection and analysis. According to the method, multi-modal data are collected and preprocessed, a weighted graph is constructed, and a global state vector is extracted; and performing distribution embedding adjustment on the global state vector, constructing a support vector machine SVM model to monitor the new energy power generation state, and collecting feedback data to optimize the support vector machine SVM model. The defect that dynamic association between data is difficult to capture in the prior art is overcome, the accuracy and the real-time performance of new energy power generation state monitoring are improved, and the comprehensiveness and the robustness of power generation state monitoring are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power generation status monitoring, and particularly to a new energy power generation status monitoring method and system. Background Art

[0002] With the in-depth promotion of the global energy transformation, new energy power stations (such as photovoltaic power stations and wind farms), as the core carriers of clean energy power generation, have been widely deployed globally. These power stations rely on natural resources (such as solar radiation and wind power) for energy conversion. However, their operating environments are highly dynamic and uncertain, facing various challenges such as meteorological condition changes, equipment aging, and abnormal operations. High-precision monitoring of the power generation status of new energy power stations is a key means to ensure power generation efficiency, improve equipment reliability, and reduce operation and maintenance costs. The power generation status monitoring of most new energy power stations relies on distributed sensor networks and data acquisition technologies, and combines traditional statistical methods or simple machine learning algorithms to achieve operation status analysis.

[0003] Existing new energy power generation status monitoring technologies still have many deficiencies in practical applications. Existing methods are often relatively rough in the fusion and processing of multi-modal data, and do not fully consider the temporal characteristics of time series data and the non-linear relationship between multi-modal data. Most existing methods do not effectively utilize efficient kernel function technologies or embedding-based machine learning models, and have limited representation capabilities for global state vectors, making it difficult to identify potential abnormal states in new energy power generation. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that existing methods are often relatively rough in the fusion and processing of multi-modal data, do not fully consider the temporal characteristics of time series data and the non-linear relationship between multi-modal data, most existing methods do not effectively utilize efficient kernel function technologies or embedding-based machine learning models, and have limited representation capabilities for global state vectors, making it difficult to identify potential abnormal states in new energy power generation.

[0006] To solve the above technical problem, the present invention provides the following technical solution: a new energy power generation status monitoring method, including: collecting comprehensive data of a first object and performing a first preprocessing.

[0007] Constructing a first task graph based on the preprocessed data and extracting vectors.

[0008] Dynamically adjusting the vectors, constructing a calculation model, and optimizing the power generation data of the first object.

[0009] Constructing a visualization interface to display the power generation data of the first object and storing the new comprehensive data of the first object generated by collection and analysis.

[0010] As a preferred solution of the new energy power generation status monitoring method described in the present invention, wherein: the first object comprehensive data is the operation-related data of the first object collected by the sensor.

[0011] As a preferred solution of the new energy power generation status monitoring method described in the present invention, wherein: the first preprocessing includes preprocessing the collected first object comprehensive data, filling in missing data, adjusting the data collection frequency to be consistent, and improving data accuracy and standardization.

[0012] As a preferred solution of the new energy power generation status monitoring method described in the present invention, wherein: constructing the first task graph includes setting edge weights and nodes to generate the first task graph.

[0013] As a preferred solution of the new energy power generation status monitoring method described in the present invention, wherein: the vector extraction includes calculating the global graph efficiency of the first task graph to generate a global vector.

[0014] As a preferred solution of the new energy power generation status monitoring method described in the present invention, wherein: the first object includes but is not limited to new energy power generation equipment, and the first object comprehensive data is multi-modal data of new energy power generation equipment.

[0015] The first preprocessing includes but is not limited to time alignment, denoising, and standardization processing.

[0016] The first task is weighted calculation, and the first task graph is a weighted graph.

[0017] As a preferred solution of the new energy power generation status monitoring method described in the present invention, wherein: the dynamic adjustment of the vector, constructing a calculation model, and optimizing the power generation data of the first object include performing distribution embedding adjustment on the global state vector, constructing a support vector machine (SVM) model to monitor the new energy power generation status, and collecting feedback data to optimize the support vector machine (SVM) model.

[0018] A new energy power generation status monitoring system, characterized in that it includes:

[0019] A data acquisition module that acquires the first object comprehensive data and performs the first preprocessing.

[0020] A vector extraction module that constructs the first task graph based on the preprocessed data and extracts vectors.

[0021] A calculation and optimization module that dynamically adjusts the vector, constructs a calculation model, and optimizes the power generation data of the first object.

[0022] A visualization module that constructs a visualization interface to display the power generation data of the first object and stores the new comprehensive data of the first object generated by collection and analysis.

[0023] A computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the method described above are implemented.

[0024] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method described above are implemented.

[0025] Advantages of the present invention: By collecting multi-modal data and performing preprocessing, constructing a weighted graph and extracting a global state vector. Adjusting the distribution embedding of the global state vector, constructing a support vector machine (SVM) model to monitor the new energy power generation status, and collecting feedback data to optimize the SVM model. It overcomes the deficiency that it is difficult to capture the dynamic association between data in the prior art, improves the accuracy and real-time performance of new energy power generation status monitoring, and enhances the comprehensiveness and robustness of power generation status monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for description in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings according to these drawings without creative efforts. Among them:

[0027] Figure 1 It is the overall flowchart of a new energy power generation status monitoring method and system provided by the first embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] To make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0029] Embodiment 1, referring to Figure 1 , which is an embodiment of the present invention, provides a new energy power generation status monitoring method, including:

[0030] S1: Collect comprehensive data of the first object and perform first preprocessing.

[0031] In the present invention, the first object comprehensive data is the multi-modal data of new energy power generation equipment. The first preprocessing is time alignment, denoising, and normalization processing.

[0032] Specifically, collecting multi-modal data and performing preprocessing means using intelligent sensors to collect multi-modal data on the power generation status in a new energy power station and performing preprocessing.

[0033] Sort the preprocessed multi-modal data in chronological order to generate a time series.

[0034] The intelligent sensors include power generation power, light, wind speed, and temperature and humidity sensors.

[0035] The multi-modal data includes power generation power, light intensity, wind speed, temperature, and humidity data.

[0036] The preprocessing includes time-aligning the multi-modal data using the Network Time Protocol, denoising the multi-modal data using windowed filtering, collaboratively filling in missing data using multi-modal time series, detecting and deleting outliers using z-score, adjusting the collection frequency of the multi-modal data to be consistent using linear interpolation, and normalizing the multi-modal data.

[0037] It should be noted that the intelligent sensors automatically collect multi-modal data on the power generation status in a new energy power station, reducing the manual intervention in data collection, improving the real-time and accuracy of the data, ensuring the efficient operation of the power generation status monitoring system. Time alignment is achieved through NTP to ensure that the multi-modal data is analyzed on a unified time axis, eliminating the impact of time asynchrony on the analysis results, improving the temporal integrity of the data, and ensuring that the generated time series has high precision. Windowed filtering effectively removes noise through local smoothing while retaining the key features of the data, improving the signal-to-noise ratio of the data and reducing the interference of noise on subsequent analysis, thereby enhancing the reliability of power generation status monitoring. By analyzing the correlation between multi-modal data, missing data is collaboratively filled in to avoid the limitations of filling in missing data with single-modal data. The z-score method quickly identifies and deletes outliers to ensure the rationality of the data distribution. The sampling frequency is adjusted through linear interpolation to enable seamless fusion of multi-modal data. Normalization processing can eliminate these differences, enabling different modal data to be compared on the same scale.

[0038] It should be noted that the first object includes but is not limited to new energy power generation equipment, and can also be energy storage equipment and transmission lines. The first preprocessing includes but is not limited to time alignment, denoising, and normalization processing, and can also be abnormal data detection and correction and data dimensionality reduction.

[0039] In an alternative embodiment of the present invention, the first object is an energy storage device. Sensor is used to collect the operation status data of the energy storage device, including state of charge (SOC) of the battery, battery voltage, battery temperature, charge and discharge power, and number of cycles. The operation log data of the energy storage system, such as abnormal event logs, is obtained in real time. The collected SOC data is standardized to ensure that the data range is between 0 and 1. Time series analysis is used to fill in the missing battery voltage and battery temperature data to ensure data integrity. Fast Fourier Transform (FFT) is applied to perform frequency domain analysis on the charge and discharge power data to remove high-frequency noise. The abnormal log events are classified and labeled to form structured data. The status data of the energy storage device is used as nodes, and the time series correlation of the battery charge and discharge power is used as the weight of the edge to generate a task graph. Topological analysis is performed on the task graph to extract key nodes and edges, representing the global efficiency of the battery state.

[0040] In an alternative embodiment of the present invention, the first object is a transmission line. Intelligent sensors are used to collect the real-time current, voltage, line loss, ambient temperature, humidity, and ice coating thickness of the transmission line. Image and video data of the transmission line are obtained through an unmanned aerial vehicle inspection system. The current, voltage, and line loss data are smoothed to remove short-term high-frequency fluctuations. Image processing technology is used to analyze the inspection images and label abnormal positions such as icing or damage points. Multimodal fusion technology is applied to combine meteorological data (temperature, humidity) and line status data (voltage, current) to form composite data. A task graph based on line nodes (such as substations, break points) is constructed, and the power transmission between lines is used as the weight of the edge. Spectral clustering is performed on the task graph to identify areas with abnormal load distribution in the transmission line.

[0041] In an alternative embodiment of the present invention, the first preprocessing may further include abnormal data detection and correction and data dimensionality reduction processing.

[0042] For abnormal data detection, statistical methods (such as box plots) are used to detect outliers in multimodal data. Long Short-Term Memory Network (LSTM) is applied to multimodal time series data to identify data mutation points and abnormal trends. For mild outliers, interpolation methods are used for smoothing to correct data anomalies. For severe outliers, historical data and prediction models (such as random forest regression) are combined for reconstruction.

[0043] The corrected data is compared with the original data to verify the correction effect. The corrected data is input into the preliminary analysis module to detect whether the analysis accuracy is improved.

[0044] Data dimensionality reduction processing uses principal component analysis (PCA) to reduce the dimensionality of high-dimensional data (such as wind speed, temperature, humidity, power generation, etc. in multi-modal data), retaining more than 80% of the main information in the data. Apply t-SNE technology to reduce the dimensionality of non-linear data features to obtain a low-dimensional representation. Compress redundant features (such as multiple highly correlated sensor data) to reduce the overhead of data storage and transmission. Use the dimensionality-reduced data as the input for task graph construction to ensure computational efficiency.

[0045] Visualize the dimensionality-reduced data in two-dimensional or three-dimensional space to verify the separability of the data and the dimensionality reduction effect. Ensure that the overall analysis result of the model is not affected after dimensionality reduction processing.

[0046] S2: Construct the first task graph based on the pre-processed data and extract vectors.

[0047] In the present invention, the first task is weighted calculation, and the first task graph is a weighted graph.

[0048] Use the Euclidean distance method to calculate the Euclidean distance ||x i - x j ||.

[0049] Use the median distance method to calculate the kernel bandwidth σ of the multi-core Gaussian kernel function. The formula is:

[0050]

[0051] where median(·) is to take the median of the distances between time series pairs, x i and x j are the i-th and j-th time series respectively, and N is the total number of time series.

[0052] The median distance, as the basis for the kernel bandwidth, can effectively reduce the risk of the influence of outliers on the kernel function. For the uneven distribution in time series (such as a significant skewness in the distribution of distance values), the median provides a more stable estimate of the central tendency. Directly using the mean or a fixed bandwidth value may have poor adaptability to time series data of different scales and is difficult to ensure robustness in the case of large samples. Calculating the kernel bandwidth based on the median can better adapt to non-uniformly distributed time series data. The growth of the sample size N may lead to changes in the distance distribution between time series. By scaling N through the logarithmic function, the bandwidth effects of large-scale samples and small-scale samples can be balanced, making the formula applicable to datasets of different scales. When the sample size is large, the bandwidth may be too large without scaling, resulting in the loss of the discrimination ability of the kernel function. When the scale is small, the scaled bandwidth can avoid overfitting caused by being too small. An incorrect bandwidth selection will directly affect the accuracy of the Gaussian kernel similarity calculation, and thus affect the performance of the entire model or algorithm. By dynamically adjusting the bandwidth by combining the median and the sample size, the kernel function can maintain high performance under different dataset distributions and scales. Using the median to calculate the distance significantly improves the robustness of the model to noise and outliers, making the similarity calculation more reliable, especially suitable for time series data with complex characteristics or a high noise level. Compared with complex optimization algorithms (such as cross-validation for bandwidth selection), this formula is simple and efficient, with low computational overhead, and can ensure the performance of the kernel function, suitable for large-scale time series similarity calculation scenarios. By combining the median of the distance distribution with logarithmic scaling, this method can adapt to changes in different time series characteristics (such as periodicity, noise level), improving the generality and robustness of the algorithm.

[0053] Calculate the similarity A of all time series using the multi-core Gaussian kernel function ij , and the formula is:

[0054]

[0055] Use the statistical quantile method to set the dynamic sparsification threshold, compare the similarity of the time series with the dynamic sparsification threshold, retain the similarity of the time series greater than the dynamic sparsification threshold, generate the dynamic association value, set it as the edge weight, use the time points as nodes, and generate a weighted graph.

[0056] Use the direct summation method to calculate the degree d of the node u in the weighted graph u , which represents the connection strength of the node u in the weighted graph, and the formula is:

[0057]

[0058] where u and v are nodes in the weighted graph, and W uv is the dynamic association value between nodes u and v.

[0059] Calculate the clustering coefficient C of node u in the weighted graph using the local connectivity weighted calculation method u , representing the local connectivity of the node, and the formula is:

[0060]

[0061] where W vw is the dynamic association value between nodes v and w, and W wu is the dynamic association value between nodes w and u.

[0062] Calculate the shortest path length L(u, v) between node u and node v using Dijkstra's algorithm.

[0063] Calculate the global graph efficiency E of the weighted graph using the shortest path inverse sum calculation method, and the formula is:

[0064]

[0065] where n is the total number of nodes.

[0066] Based on the degree d u of node u and the clustering coefficient C u , use the weighted average method to calculate the global node degree d glo and the global clustering coefficient C bal respectively.

[0067] Generate the global state vector Z from the global node degree d glo of the weighted graph, the global clustering coefficient C bal and the global graph efficiency E, where Z = [d glo , C bal , E].[[]END]

[0068] By calculating the Euclidean distance between pairs of time series, it provides a basis for constructing the edge weights in the weighted graph, effectively quantifying the numerical differences in time series, ensuring the accuracy of similarity calculation, and at the same time providing accurate input for subsequent kernel function calculation. The median distance method can provide a stable bandwidth estimate when there are outliers in the sample distribution, avoid the interference of outliers on kernel function calculation, and improve the robustness and reliability of time series similarity calculation. The multi-kernel Gaussian kernel function can capture the non-linear similarity between time series by combining multiple kernel functions, while retaining the overall trend and local features, significantly improving the accuracy of similarity calculation, enabling the edge weights of the weighted graph to more truly reflect the dynamic relationship between time series, making the weighted graph retain a strong correlation value, while removing invalid or weakly associated edges, enhancing the sparsity of the weighted graph, improving the calculation efficiency of subsequent graph algorithms, and at the same time retaining important structural information. By summing and calculating the local connectivity, the degree and clustering coefficient of the nodes are obtained, comprehensively reflecting the connection characteristics and local association strength at each time point. The node features can effectively describe the role of time points in the global graph, providing important support for generating the global state vector. By using the Dijkstra algorithm to quickly calculate the shortest path between nodes, combining the inverse of the shortest path and calculating the global graph efficiency, the global connectivity of the time series is reflected. By calculating the global node degree and clustering coefficient through weighted averaging and combining with the global graph efficiency, the global state vector is generated. The global state vector condenses the global information of the weighted graph in a low-dimensional space, providing efficient input for subsequent model training and state monitoring.

[0069] S3: Dynamically adjust the vector, construct a calculation model, and optimize the power generation data of the first object.

[0070] Perform distribution embedding adjustment on the global state vector, construct a support vector machine (SVM) model to monitor the new energy power generation status, and collect feedback data to optimize the SVM model.

[0071] Specifically, performing distribution embedding adjustment on the global state vector and constructing an SVM model to monitor the new energy power generation status means collecting historical multi-modal data for preprocessing and extracting the historical global state vector Z 1 。

[0072] Use the Wasserstein distance to calculate the Wasserstein distance W(Z, Z 1 ) between the global state vector Z and the historical global state vector Z 1 ), and the formula is:

[0073]

[0074] where Π(Z, Z 1 ) is the Wasserstein distance between the global state vector Z and the historical global state vector Z 1The set of joint distributions, ||z - Z 1 || is the Euclidean distance between the global state vector Z and the historical global state vector Z 1 , dγ(Z, Z 1 ) is the differential element under the γ distribution, γ is a distribution in the set of joint distributions Π(Z, Z 1 ), and inf is to take the distribution that minimizes the Wasserstein distance among all joint distributions.

[0075] Set the sliding window size using autocorrelation analysis, and smooth the Wasserstein distance W(Z, Z 1 ) using the distribution change of the sliding window to obtain the smoothed Wasserstein distance W amo .

[0076] Set the maximum allowable value W of the Wasserstein distance through historical distribution statistics max , and calculate the dynamically adjusted step size η. The formula is:

[0077]

[0078] where ΔW amo is the change in the smoothed Wasserstein distance.

[0079] Calculate the gradient of the Wasserstein distance with respect to the global state vector Z The formula is:

[0080]

[0081] where is the partial derivative of the Wasserstein distance with respect to the historical global state vector Z 1 .

[0082] Perform distribution embedding adjustment on the global state vector Z to obtain the distribution embedding feature vector Z'. The formula is:

[0083]

[0084] Collect labeled training multimodal data through the UCI machine learning library for preprocessing, extract the training distribution embedding feature vectors, and generate a training set.

[0085] Calculate the mean of the training distribution embedding feature vector Z * as the reference feature vector Z 3 , and randomly select a subset from the training set as the independent distribution embedding feature vector Z 2 for support vector machine SVM model training.

[0086] Construct a Support Vector Machine (SVM) model, including an input layer and a kernel function output layer.

[0087] Define the input layer as the distribution embedding feature vector Z'.

[0088] Calculate the kernel bandwidth σ' of the RBF kernel function based on the sample distribution density to control the influence range of the Euclidean distance term.

[0089] Use the RBF (Radial Basis Function) as the RBF kernel function K(Z 2 ,Z 3 ), and the formula is:

[0090]

[0091] where δ is the weight parameter, and ||Z 2 -Z 3 || 2 is the square of the Euclidean distance between the independent distribution embedding feature vector Z 2 and the reference feature vector Z 3 .

[0092] The RBF kernel function has a strong non - linear mapping ability, which can project the original time - series data into a high - dimensional feature space, making it easier to distinguish complex time - series patterns in the high - dimensional space. The RBF kernel processes the Euclidean distance in an exponential form, ensuring that smaller distances contribute larger similarity values and larger distances quickly decay to near zero, thus strengthening the local similarity features. Cosine similarity measures the directional matching between time series rather than just the proximity of numerical distributions. This correction term can complement the limitation of the RBF kernel function based only on distance. The weight parameter provides control over the influence range of the correction term, enabling the algorithm to select a more suitable kernel function weight allocation according to specific problems. Without introducing the correction term, the RBF kernel function may ignore the directional information of time series and perform poorly in some specific application scenarios (such as periodic signal analysis). The introduction of the correction term avoids the bias caused by simply relying on distance or direction, improves the applicability of the kernel function in complex scenarios, and realizes the effective fusion of distance features and direction features, making it applicable to a wider range of time - series analysis scenarios.

[0093] The decision output layer, and the formula is:

[0094] f(Z 2 )=sgn(K(Z 2 ,Z 3 )+b)

[0095] where f(Z 2 ) is the output state label, and b is the bias term.

[0096] Train the support vector machine (SVM) model using the training set, and use the loss function and Adam optimizer to iteratively optimize the model parameters.

[0097] Input the distribution-embedded feature vector Z' into the trained SVM model to obtain state labels, including -1 and 1.

[0098] If the state label is 1, it is judged as the normal state, and the multimodal data is continuously monitored. If the state label is -1, it is judged as the abnormal state, and a warning is issued and the maintenance personnel are reminded by email.

[0099] Through the preprocessing of historical data and the extraction of global state vectors, a stable foundation is provided for subsequent analysis. The distance metric measures the deviation between the real-time state and the historical state, providing a scientific basis for anomaly judgment, effectively capturing the subtle changes in the global state distribution, enhancing the sensitivity and reliability of anomaly detection. By setting the sliding window size through autocorrelation analysis, the smoothing effect is further improved, and the interference of short-term fluctuations of the Wasserstein distance on the results is reduced, ensuring a more stable judgment of state changes. During the distribution embedding process, the vector features are dynamically adjusted to generate the embedded feature vector, simplifying the feature expression and reducing the computational cost, providing optimized input data for the training and inference of the SVM model. By defining the input layer as the distribution-embedded feature vector and combining with the RBF kernel function, the classification ability of the model for non-linearly distributed data is enhanced, efficiently processing complex multimodal features, accurately identifying the normal and abnormal modes of new energy power generation status. By dynamically adjusting the bandwidth according to the sample distribution density, the kernel function is ensured to adapt to different feature distributions, improving the adaptability of the kernel function, avoiding underfitting caused by too large bandwidth or overfitting caused by too small bandwidth. The optimization method not only accelerates the model convergence but also improves the training stability, significantly enhancing the model training efficiency, ensuring the generalization ability of the model in complex new energy power generation scenarios, realizing real-time anomaly monitoring and automatic alarm, improving the safety and operation efficiency of the power generation system, and reducing the economic losses caused by the failure to detect abnormal states in a timely manner.

[0100] Furthermore, collecting feedback data to optimize the SVM model includes

[0101] Collect the feedback data of the maintenance personnel and preprocess it, and use the confusion matrix analysis method to calculate the false alarm rate of the SVM model.

[0102] Construct an optimization objective to minimize the error rate, calculate the weight parameters of the RBF kernel function using the feedback data, and adjust the weight parameters of the RBF kernel function according to the gradient descent method.

[0103] Use the fixed iteration limit method to set the termination threshold, and stop the iteration when the number of iterations reaches the termination threshold.

[0104] Substitute the updated weight parameters of the RBF kernel function into the support vector machine (SVM) model and continue to monitor the new energy power generation status.

[0105] By cleaning, organizing, and analyzing this feedback data, the deficiencies of the model can be effectively discovered, enhancing the pertinence of the model optimization process, enabling the model to gradually adapt to the actual operating environment. By using the feedback data to improve the practicality and accuracy of the monitoring model, the problem of the SVM model in monitoring can be accurately located by calculating the false alarm rate, accurately quantifying the frequency of misclassification of the model, providing precise data support for subsequent optimization. It helps to improve the model performance, reduce the interference of false alarms on the system operation. By clearly taking the minimization of the error rate as the optimization goal, the present invention can systematically optimize the model performance, reduce false alarms and missed alarms, enhance the model's detection ability for abnormal states, reduce the risk of normal states being misjudged as abnormal, and enhance the reliability of the system and user trust. By using the feedback data to recalculate the weight parameters, the kernel function can dynamically adapt to the new data distribution, enhancing the model's adaptability to complex data patterns, and improving the accuracy and real-time performance of monitoring. Through feedback optimization, the self-evolution ability of the model is realized. The gradient descent method is used to optimize the kernel function parameters to ensure the convergence and efficiency of the weight adjustment process, significantly reducing the time and resource consumption of the optimization calculation, while improving the accuracy of the optimization result, providing an efficient optimization tool for the new energy power generation status monitoring. By setting the maximum number of iterations or error threshold, the optimization process can stop in time after meeting the conditions, avoiding meaningless calculations and preventing over-optimization from leading to model complexity or overfitting. It improves the utilization rate of computing resources, ensuring the efficiency and stability of the optimization process. The optimized weight parameters of the kernel function are dynamically integrated into the support vector machine model, enhancing the model's monitoring ability for the new energy power generation status.

[0106] S4: Construct a visual interface to display the power generation data of the first object, and store the new comprehensive data of the first object generated by collection and analysis.

[0107] Specifically, use the front-end technology Chart.js to construct a visual interface to display the new energy power generation status, and use HTML and CSS to construct the layout of the visual interface.

[0108] Allow users who have passed real-name verification to access.

[0109] The visual interface layout includes an area for displaying the new energy power generation status and an interaction area.

[0110] Chart.js provides efficient data visualization capabilities, which can dynamically display the new energy power generation status data in various chart forms, including key indicators such as real-time power generation power, light intensity, and wind speed. The data display is intuitive and clear, facilitating users to quickly understand the power generation status and trends. It supports the dynamic update function, which can reflect the changes in the power generation status in real time, improving the monitoring efficiency of the system. Through the combination of HTML and CSS, the visualization interface has good readability and aesthetics. It enhances the user experience, making it easy for users to quickly find the information they need. By restricting access rights through real-name verification, it ensures the security and privacy of the system data. Only authorized users can access the new energy power generation status data, avoiding the leakage of sensitive information. The display area is used to dynamically display the key data of the new energy power generation status. The interaction area allows users to adjust the data display method (such as switching the time range, selecting the data type, etc.) or query the data for a specific time period. The interaction area also allows users to set a specific time period or parameters to query the corresponding data chart. The real-time update function ensures the timeliness of the data, helping users quickly discover anomalies or trends. The interaction function enhances the intelligence level of the system, enabling users to deeply explore the data value. The data trend analysis provides decision-making support for users. The anomaly warning function helps users quickly discover problems and reduce system risks.

[0111] Furthermore, storing the multi-modal data generated by collection and analysis means storing the collected multi-modal data and the new energy power generation status generated by analysis in the central database. The central database sorts the data in chronological order and marks the corresponding tags. At the same time, it synchronously backs up the collected multi-modal data and the new energy power generation status generated by analysis to the cloud, and regularly detects the integrity of the backup data.

[0112] By storing the collected multi-modal data and the new energy power generation status obtained by analysis in the central database, a centralized data storage and management platform is formed. The chronological sorting and tag marking functions of the database facilitate subsequent data retrieval and analysis. The automatic synchronization function of the backup data further improves the automation level and reliability of the system. The method of regular detection helps to timely discover and repair possible data anomalies. The analysis results are directly associated with the original data, forming a closed-loop management mode of "data - analysis - result". The data analysis and modeling provide a unified data source, simplifying the process. The traceability of the analysis results enhances the transparency and user trust of the system. The associated management of data and results improves the intelligence level of the system, supporting more advanced decision-making support and prediction functions. The data is sorted in chronological order and tagged, enabling the system to efficiently handle time series analysis tasks. The tag marking helps to quickly locate anomalies or special events, providing support for fault diagnosis and warning.

[0113] The computer device can be a server. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the data cluster data of the power monitoring system. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a new energy power generation status monitoring method.

[0114] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0115] Embodiment 2, an embodiment of the present invention, provides a new energy power generation status monitoring method and system. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0116] In order to verify the innovation of the new energy power generation status monitoring method of our invention in terms of improving monitoring accuracy, real-time performance, and system stability, the following test is designed. This test takes a certain photovoltaic power station as the test scenario and tests the key performance indicators of power generation status monitoring, including the accuracy of data collection, the abnormal detection ability of the model, and the accurate recognition ability of the power generation status trend. The comparison method is a monitoring technology based on traditional statistics and simple machine learning algorithms.

[0117] Existing technology experimental process:

[0118] Traditional sensors are used to collect single-modal data such as the power generation of a photovoltaic power station, light intensity, wind speed, temperature, and humidity. The collected data is simply denoised, but not time-aligned or standardized. A linear regression model is used to analyze the trend of the power generation status, without involving the fusion and embedding processing of multi-modal data. Static charts are generated, and the abnormal status is monitored manually, without real-time warning function.

[0119] The experimental process of our invention is as follows:

[0120] Data such as power generation, light intensity, wind speed, temperature, and humidity are collected through multi-modal intelligent sensors. Standardization processing is carried out by using methods such as time alignment, denoising, z-score anomaly detection, and linear interpolation, and high-frequency noise is removed. A weighted graph is generated from time series data, the dynamic correlation value is calculated through a multi-kernel Gaussian kernel function, the global state vector is extracted, and the non-linear relationship between data is retained. A support vector machine (SVM) model is constructed, the feature vector is adjusted by using distribution embedding, and the model weight is dynamically optimized through feedback data to improve the abnormal status detection ability. Chart.js is used to generate a dynamic visualization interface to display the trend of the power generation status and abnormal warning; the data is stored in a central database and backed up regularly.

[0121] The experimental results are shown in Table 1.

[0122] Table 1 Experimental Results

[0123]

[0124]

[0125] Through the comparison of the tables, the advantages of our invention in key indicators can be clearly seen:

[0126] Under various operating conditions, through the fusion and embedding processing of multi-modal data, our invention makes the numerical value of power monitoring closer to the actual value. For example, when the light is insufficient, the existing technology only monitors 95 kW, while our invention monitors 96.2 kW, and the error is significantly reduced. Especially in rainy and high-wind environments, our invention has stronger adaptability to complex scenarios.

[0127] The detection rate of abnormal status of our invention is much higher than that of the existing technology. In the test of short-term equipment failure, the existing technology did not detect any abnormality at all, while our invention achieved a detection rate of 90% through distribution embedding adjustment and SVM model optimization.

[0128] Our invention can capture the non-linear relationship between multi-modal data by using time alignment, standardization processing, and multi-kernel Gaussian kernel function, and shows higher robustness in complex environments. The existing technology has limited ability to process noise and non-linear features in data, resulting in an abnormal detection rate of only 30%-75%.

[0129] Our invention displays the power generation status trend through a dynamic visualization interface, gives real-time warnings of anomalies, and improves the intelligence level of the monitoring system. The prior art only generates static charts and requires manual analysis of anomalies, resulting in low efficiency.

[0130] In summary, our invention significantly overcomes the deficiencies of the prior art in terms of monitoring accuracy, anomaly detection, and system intelligence. Through comprehensively innovative technical means, it realizes more efficient and accurate monitoring of new energy power generation status.

[0131] Example 3. Among the above examples, it is an example of the present invention, including a new energy power generation status monitoring system, specifically:

[0132] A data acquisition module that acquires comprehensive data of the first object and performs first preprocessing;

[0133] A vector extraction module that constructs a first task graph based on the preprocessed data and extracts vectors;

[0134] A calculation and optimization module that dynamically adjusts the vectors, constructs a calculation model, and optimizes the power generation data of the first object;

[0135] A visualization module that constructs a visualization interface to display the power generation data of the first object and stores the new comprehensive data of the first object generated by collection, analysis.

[0136] It should be noted that the above examples are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred examples, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for monitoring the state of new energy power generation, characterized in that: include: Collecting comprehensive data of the first object and performing first pre-processing; Construct a first task graph based on the pre-processed data and extract vectors; Dynamically adjust the vector, build a calculation model, and optimize the power generation data of the first object; Construct a visualization interface to display the power generation data of the first object, and store the new first object comprehensive data generated by collection and analysis.

2. The method for monitoring the state of new energy power generation according to claim 1, characterized in that: The first object comprehensive data is first object operation related data collected by the sensor.

3. The method for monitoring the state of new energy power generation according to claim 2, characterized in that: The first pre-processing includes pre-processing the collected comprehensive data of the first object, filling in missing data, adjusting the data collection frequency to keep it consistent, and improving data accuracy and standardization.

4. The method for monitoring the state of new energy power generation according to claim 3, characterized in that: The constructing the first task graph includes setting edge weights and nodes to generate the first task graph.

5. The method for monitoring the state of new energy power generation according to claim 4, characterized in that: The extracting vector includes calculating the global graph efficiency of the first task graph and generating a global vector.

6. The method for monitoring the state of new energy power generation according to claim 5, characterized in that: The first object includes but is not limited to new energy power generation equipment, and the first object comprehensive data is multimodal data of new energy power generation equipment; The first pre-processing includes but is not limited to time alignment, denoising and normalization processing; The first task is weighted calculation, and the first task graph is a weighted graph.

7. The method for monitoring the state of new energy power generation according to claim 6, characterized in that: The dynamically adjusting the vector, constructing the calculation model, and optimizing the power generation data of the first object include: distributing and embedding the global state vector, constructing a support vector machine (SVM) model to monitor the power generation state of new energy, and collecting feedback data to optimize the support vector machine (SVM) model.

8. A new energy power generation status monitoring system using the method according to any one of claims 1 to 7, characterized in that: A data collection module collects comprehensive data of the first object and performs a first pre-processing; A vector extraction module constructs a first task graph according to the pre-processed data and extracts vectors; A calculation optimization module dynamically adjusts the vector, builds a calculation model, and optimizes the power generation data of the first object; The visualization module constructs a visualization interface to display the power generation data of the first object, and stores the new first object comprehensive data generated by collection and analysis.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images