Photovoltaic equipment communication abnormity diagnosis method, medium and system

Abstract

The invention provides a photovoltaic equipment communication abnormity diagnosis method, medium and system, and belongs to the technical field of communication abnormity diagnosis, and the method comprises the steps: firstly carrying out the business boundary division through determining a communication problem domain and a description language thereof, and obtaining a boundary context; secondly, acquiring communication real-time data by using a multi-channel acquisition system, performing variation decomposition on the data to obtain a stable component and a variation component, and establishing a mapping function between the stable component and the variation component; a parameter stability coefficient matrix is calculated based on the change relation, and historical data is acquired and filtered to form a sample set; various preset characteristic parameters are acquired, and communication delay, quality and frequency characteristics are calculated by using the optimization equation set; training a random inactivation neural network model by using a historical sample, and calculating an abnormal probability and an uncertainty evaluation value through multiple forward propagation; and finally, performing clustering analysis based on the data to obtain a communication anomaly type and determine a fault diagnosis result.

Description

Technical Field

[0001] The present invention belongs to the technical field of communication anomaly diagnosis, and in particular, relates to a method, medium and system for diagnosing communication anomaly of photovoltaic equipment. Background Art

[0002] With the rapid development of photovoltaic power generation technology, photovoltaic equipment has been widely used around the world. As an important carrier of clean energy power generation, the safe and stable operation of photovoltaic power stations is directly related to the power supply quality and economic benefits of the power grid. During the operation of photovoltaic power stations, the communication system undertakes key tasks such as data transmission, monitoring command issuance, and equipment status collection. However, due to the characteristics of photovoltaic power stations, which are generally distributed over a wide range, with a large number of equipment and a complex environment, the communication system often faces various abnormal situations.

[0003] At present, the diagnosis of photovoltaic equipment communication anomalies mainly adopts the following technical solutions: rule-based anomaly detection method, statistical analysis method and traditional machine learning method. The rule-based anomaly detection method identifies communication anomalies through pre-set thresholds and judgment rules. This method is simple to implement, but it is difficult to adapt to complex and changeable communication environments, and requires a lot of manual experience to set rules. The statistical analysis method uses the statistical characteristics of historical data to judge communication anomalies. Although it has a certain mathematical foundation, it has high requirements on data quality and is difficult to capture the dynamic characteristics of communication anomalies. Traditional machine learning methods such as support vector machines (SVM) and decision trees can automatically learn data features, but they are not effective when processing high-dimensional and nonlinear communication data.

[0004] In practical applications, the existing technologies have the following outstanding problems: First, communication anomalies manifest in various forms, including communication delays, packet loss, signal quality degradation, etc., and a single detection method is difficult to fully cover various abnormal situations; second, the operating environment of photovoltaic equipment is complex, and factors such as climatic conditions and electromagnetic interference will affect the communication quality. It is difficult for existing methods to effectively distinguish between normal fluctuations and real anomalies; third, communication anomalies often have gradual characteristics. In the early stage of the anomaly, it is difficult for traditional methods to detect the problem in time; fourth, the existing technologies generally lack credibility assessment of the diagnostic results, which is prone to false alarms or omissions.

[0005] In addition, the data generated by the photovoltaic equipment communication system has obvious time series characteristics and multi-dimensional attributes, and there are complex correlations between these data. When processing such data, the existing technology often uses simple data preprocessing methods, which fails to fully utilize the inherent correlation between the data, resulting in unsatisfactory diagnostic results. Summary of the invention

[0006] In view of this, the present invention provides a photovoltaic equipment communication anomaly diagnosis method, medium and system, which can solve the technical problem that the prior art is difficult to fully utilize the inherent relationship between data, resulting in unsatisfactory diagnosis effect.

[0007] The present invention is achieved in that:

[0008] A first aspect of the present invention provides a method for diagnosing abnormal communication of a photovoltaic device, comprising the following steps:

[0009] S10, obtaining a communication problem domain of the photovoltaic device, determining a description language of the communication problem domain, dividing the boundaries of the communication service domain based on the description language, and obtaining a boundary context of the communication service domain;

[0010] S20, using multiple data acquisition channels to collect communication real-time data from the photovoltaic device, wherein one of the data acquisition channels corresponds to one type of the communication real-time data, and the multiple data acquisition channels are used to collect multiple types of the communication real-time data;

[0011] S30, performing variation decomposition calculation on the real-time communication data to obtain a stable component of the communication data and a variation component of the communication data;

[0012] S40, establishing a mapping function between the stable component of the communication data and the variable component of the communication data, recorded as a communication variable relationship, and establishing a communication parameter variable coefficient matrix according to the communication variable relationship;

[0013] S50, calculating a communication parameter stability coefficient matrix based on the communication parameter variation coefficient matrix; obtaining historical communication data of the photovoltaic device, filtering the historical communication data using a data filtering rule, and generating a historical feature sample set;

[0014] S60, obtaining a preset feature weight coefficient, a preset feature threshold upper limit, a preset feature threshold lower limit, and a preset feature importance coefficient;

[0015] S70, optimizing and calculating the communication parameter stability coefficient matrix using a communication parameter optimization equation group to obtain communication delay characteristics, communication quality characteristics, and communication frequency characteristics;

[0016] S80, training a random loss neural network model based on the historical feature sample set, and calculating abnormal probability values ​​and uncertainty assessment values ​​of the communication delay feature, the communication quality feature, and the communication frequency feature through multiple forward propagations;

[0017] S90, performing cluster analysis based on the abnormal probability value, the uncertainty evaluation value, and the communication parameter stability coefficient matrix to obtain a communication abnormality type, and determining a fault diagnosis result according to the communication abnormality type.

[0018] On the basis of the above technical solution, a photovoltaic equipment communication abnormality diagnosis method of the present invention can also be improved as follows:

[0019] Wherein, the step S10 specifically includes:

[0020] Step 101, obtaining a communication problem domain of a photovoltaic device;

[0021] Step 102: Determine the description language of the communication problem domain;

[0022] Step 103: Analyze the characteristics of the communication service field according to the description language;

[0023] Step 104: dividing the communication service area into boundaries based on the characteristics;

[0024] Step 105: determining the boundary context of the communication service domain according to the boundary division result;

[0025] The step S20 specifically includes:

[0026] Step 201, setting a plurality of data acquisition channels, each of the data acquisition channels corresponding to a type of communication real-time data;

[0027] Step 202: collecting communication real-time data from the photovoltaic device through the multiple data collection channels;

[0028] Step 203: Classify and store the real-time communication data collected by the multiple data collection channels.

[0029] Furthermore, the step S30 specifically includes:

[0030] Step 301: identifying extreme points based on real-time communication data;

[0031] Step 302, constructing an upper envelope and a lower envelope for the extreme points using cubic spline interpolation;

[0032] Step 303, calculating the envelope mean of the upper envelope and the lower envelope;

[0033] Step 304: subtract the envelope mean value from the communication real-time data to obtain a first component;

[0034] Step 305: Perform intrinsic mode decomposition on the communication real-time data to obtain multiple intrinsic mode functions and residual functions, wherein the residual function serves as a stable component of the communication data, and the superposition of the multiple intrinsic mode functions serves as a variable component of the communication data.

[0035] Furthermore, the step S40 specifically includes:

[0036] Step 401, calculating the correlation items between stable components of communication data of different dimensions;

[0037] Step 402: Calculate the correlation items between the communication data change components of different dimensions;

[0038] Step 403, calculating the correlation term between the gradient of the stable component of the communication data and the gradient of the variable component of the communication data;

[0039] Step 404, introducing a time decay term and an error term;

[0040] Step 405: Form a communication change relationship by weighted combination of the association term, the time decay term and the error term;

[0041] Step 406: construct a communication parameter variation coefficient matrix based on the communication variation relationship.

[0042] Furthermore, the step S50 specifically includes:

[0043] Step 501, calculating the norm of the communication parameter variation coefficient matrix;

[0044] Step 502: normalize the communication parameter variation coefficient matrix;

[0045] Step 503: introducing a time weight coefficient and a time scale parameter;

[0046] Step 504: construct a time decay function;

[0047] Step 505: Combining the normalized matrix with the time decay function to obtain a communication parameter stability coefficient matrix;

[0048] Step 506: Acquire historical communication data of the photovoltaic device;

[0049] Step 507: pre-process the historical communication data according to preset data filtering rules;

[0050] Step 508: extracting characteristic parameters from the pre-processed historical communication data;

[0051] Step 509: Generate a historical feature sample set based on the feature parameters.

[0052] Furthermore, the step S70 specifically includes:

[0053] Step 701: Input the characteristic weight coefficient and the communication parameter stability coefficient matrix into the characteristic mapping equation, which contains linear terms, nonlinear mapping terms and time derivative terms, and output the initial mapping value of the communication characteristic;

[0054] Step 702: input the communication feature initial mapping value and the feature threshold upper limit and feature threshold lower limit into a feature constraint equation, the feature constraint equation uses a sigmoid function to perform feature constraints, and outputs a constrained communication feature value;

[0055] Step 703: input the constrained communication eigenvalue and the characteristic importance coefficient into a characteristic balance equation, where the characteristic balance equation includes a linear weight term, a nonlinear weight term and an acceleration weight term, and output the balanced communication eigenvalue;

[0056] Step 704: input the balanced communication characteristic value and the historical characteristic sample set into a characteristic optimization equation, which contains a spatial regularization term, a time derivative term, a historical sample constraint term, and a weighted integration term, and outputs communication delay characteristics, communication quality characteristics, and communication frequency characteristics.

[0057] Furthermore, the step S80 specifically includes:

[0058] Step 801, constructing a random dropout neural network model;

[0059] Step 802: training the random dropout neural network model based on the historical feature sample set;

[0060] Step 803, performing multiple forward propagation calculations using the random dropout neural network model;

[0061] Step 804, respectively calculating the abnormal probability values ​​of the communication delay feature, the communication quality feature, and the communication frequency feature;

[0062] Step 805: Calculate uncertainty evaluation values ​​of the communication delay characteristics, the communication quality characteristics, and the communication frequency characteristics respectively.

[0063] Furthermore, the step S90 specifically includes:

[0064] Step 901: taking the abnormal probability value, the uncertainty assessment value and the communication parameter stability coefficient matrix as clustering features;

[0065] Step 902: performing cluster analysis on the cluster features;

[0066] Step 903: Determine the type of communication anomaly according to the cluster analysis result;

[0067] Step 904: Determine a fault diagnosis result of the photovoltaic device based on the communication abnormality type.

[0068] The communication parameter optimization equation group includes a characteristic mapping equation, a characteristic constraint equation, a characteristic balance equation and a characteristic optimization equation:

[0069] The characteristic mapping equation input includes the values ​​of each element of the communication parameter stability coefficient matrix and the preset characteristic weight coefficient, and outputs the initial mapping value of the communication characteristic;

[0070] The characteristic constraint equation input includes the initial mapping value of the communication characteristic, the preset characteristic threshold upper limit and the preset characteristic threshold lower limit, and outputs the communication characteristic value after constraint;

[0071] The characteristic balance equation input includes the constrained communication characteristic value and the preset characteristic importance coefficient, and outputs the balanced communication characteristic value;

[0072] The characteristic optimization equation input includes the balanced communication characteristic value and the historical characteristic sample set, and outputs the communication delay characteristic, the communication quality characteristic and the communication frequency characteristic.

[0073] The formulas or equations involved in the present invention are described in detail below:

[0074] 1. The specific expression of the change decomposition calculation in S30 is as follows:

[0075]

[0076] Where, Y(t) is the real-time communication data; c i (t) is the i-th intrinsic mode function, representing the fluctuation components of different frequency scales; r(t) is the residual function; n is the decomposition order; t is the time variable.

[0077] Where: stable component of communication data S(t) = r(t); variable component of communication data

[0078] Parameter acquisition method:

[0079] (1) Identify the extreme points of Y(t);

[0080] (2) Use cubic spline interpolation to construct the upper and lower envelopes e max (t),e min (t);

[0081] (3) Calculate the envelope mean: m 1 (t) = [e max (t)+emin (t)] / 2;

[0082] (4) Extraction component: h 1 (t) = Y(t) - m 1 (t).

[0083] 2. The specific expression of the communication change relationship in S40 is as follows:

[0084]

[0085] Where M ij is the element of the communication parameter variation coefficient matrix; S i (t),S j (t) is the stable component of communication data in different dimensions; V i (t),V j (t) is the communication data variation component of different dimensions; α 1 ,α 2 ,α 3 ,α 4 is the weight coefficient; λ is the time attenuation coefficient; t 0 is the reference time; ∈ is the error term, ranging from 0.01 to 0.05.

[0086] 3. The communication parameter stability coefficient matrix in S50 is calculated as follows:

[0087]

[0088] in:

[0089] Where K is the communication parameter stability coefficient matrix; M ij is the element of the communication parameter variation coefficient matrix; β is the time weight coefficient; μ is the time scale parameter.

[0090] 4. The specific expression of the characteristic mapping equation is as follows:

[0091]

[0092] Where F is the communication feature mapping matrix; W is the feature weight matrix; K is the communication parameter stability coefficient matrix; φ(K) is the nonlinear mapping function; γ is the time derivative weight; ∈ 1 is the mapping error matrix.

[0093] 5. The specific expression of the characteristic constraint equation is as follows:

[0094]

[0095] In the formula, G ijis the element of the constrained feature matrix; σ(x) is the sigmoid function; θ max ,θ min are the upper and lower thresholds of the feature respectively; 2 is the constrained error term.

[0096] 6. The specific expression of the characteristic balance equation is as follows:

[0097]

[0098] Where H is the balanced feature matrix; P is the linear weight matrix; Q is the nonlinear weight matrix; ⊙ represents the Hadamard product; R is the acceleration weight matrix; ∈ 3 is the balance error term.

[0099] 7. The specific expression of the characteristic optimization equation is as follows:

[0100]

[0101] Where O is the optimized feature matrix, which includes communication delay features, communication quality features, and communication frequency features; is the Laplace operator; H * For historical characteristic samples; is the i-th historical sample; ω i is the sample weight; 1 ,λ 2 ,λ 3 ,λ 4 is the optimization coefficient; ∈ 4 is the optimization error term.

[0102] 8. The forward propagation calculation of the random dropout neural network in S80 is expressed as:

[0103] Z (l) =W (l) A (l-1) D (l) +b (l) ;

[0104] A (l) =f(Z (l) );

[0105] In the formula, Z (l) is the weighted input of the lth layer; W (l) is the weight matrix; A (l-1) is the activation value of the previous layer; D (l) is the random dropout matrix; b (l) is the bias vector; f is the activation function.

[0106] Equation principle explanation:

[0107] 1. The feature mapping equation introduces nonlinear mapping and time derivative terms to enhance feature extraction capabilities;

[0108] 2. The feature constraint equation uses the sigmoid function to implement soft constraints to avoid information loss caused by hard threshold truncation;

[0109] 3. The characteristic balance equation takes into account linear terms, nonlinear terms and acceleration terms to improve the dynamic balance of the characteristics;

[0110] 4. The feature optimization equation includes spatial regularization terms, time derivative terms, historical sample constraint terms and weighted integration terms to achieve multi-dimensional optimization;

[0111] 5. Random Dropout Neural Network improves the generalization ability of the model by randomly disconnecting connections.

[0112] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed in a computer, they are used to execute the above-mentioned photovoltaic device communication abnormality diagnosis method.

[0113] A third aspect of the present invention provides a photovoltaic equipment communication anomaly diagnosis system, which includes the above-mentioned computer-readable storage medium.

[0114] Compared with the prior art, the photovoltaic equipment communication abnormality diagnosis method, medium and system provided by the present invention have the following beneficial effects:

[0115] 1. Multi-dimensional perception of communication problems. This method collects real-time communication data of photovoltaic equipment through multiple channels, including delay, quality, frequency and other dimensions, and can fully perceive the working status of the communication system, laying the foundation for subsequent fault diagnosis. In contrast, existing monitoring methods based on overall performance often have difficulty in accurately locating the root cause of communication problems.

[0116] 2. Deeply explore communication features. This method uses advanced signal processing and machine learning algorithms, such as intrinsic mode decomposition, stability analysis, neural network optimization, etc., to deeply explore the time-varying, nonlinear and historical dependence characteristics of communication data, thereby building a more accurate communication anomaly diagnosis model. In contrast, the diagnosis method based on empirical rules cannot fully characterize the complex characteristics of communication failures.

[0117] 3. Improve the accuracy and reliability of diagnosis. This method integrates multi-dimensional features, combines abnormal probability, uncertainty assessment and stability analysis, etc., and can more objectively and accurately identify the type of communication anomalies, greatly improving the reliability of fault diagnosis. In contrast, existing methods based on protocol analysis are easily affected by the limitations of predetermined fault modes.

[0118] 4. Targeted fault location and resolution. This method can clearly diagnose the specific type of communication anomaly and provide targeted guidance for subsequent fault resolution. Compared with the general results given by previous empirical diagnosis methods, this method can help maintenance personnel locate and resolve problems more quickly and efficiently.

[0119] In summary, the photovoltaic equipment communication anomaly diagnosis method proposed in the present invention makes full use of the inherent laws of communication data and adopts advanced signal processing and machine learning technologies to achieve accurate identification and positioning of communication system faults, solving the technical problem that the existing technology is difficult to fully utilize the inherent correlation between data, resulting in unsatisfactory diagnostic effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0120] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0121] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0122] like Figure 1 As shown, it is a flow chart of a photovoltaic device communication abnormality diagnosis method provided by the first aspect of the present invention. The method comprises the following steps:

[0123] S10, obtaining a communication problem domain of the photovoltaic device, determining a description language of the communication problem domain, dividing the boundaries of the communication business domain based on the description language, and obtaining a boundary context of the communication business domain;

[0124] S20, using multiple data acquisition channels to collect communication real-time data from the photovoltaic device, where one data acquisition channel corresponds to one type of communication real-time data, and multiple data acquisition channels are used to collect multiple types of communication real-time data;

[0125] S30, performing variation decomposition calculation on the real-time communication data to obtain a stable component of the communication data and a variation component of the communication data;

[0126] S40, establishing a mapping function between a stable component of communication data and a variable component of communication data, recorded as a communication variable relationship, and establishing a communication parameter variable coefficient matrix according to the communication variable relationship;

[0127] S50, calculating the communication parameter stability coefficient matrix based on the communication parameter variation coefficient matrix; obtaining historical communication data of the photovoltaic device, filtering the historical communication data using data filtering rules, and generating a historical feature sample set;

[0128] S60, obtaining a preset feature weight coefficient, a preset feature threshold upper limit, a preset feature threshold lower limit, and a preset feature importance coefficient;

[0129] S70, optimizing and calculating the communication parameter stability coefficient matrix using the communication parameter optimization equation group to obtain communication delay characteristics, communication quality characteristics, and communication frequency characteristics;

[0130] S80, training a random loss neural network model based on a historical feature sample set, and calculating abnormal probability values ​​and uncertainty assessment values ​​of communication delay features, communication quality features, and communication frequency features through multiple forward propagations;

[0131] S90, performing cluster analysis based on the abnormal probability value, the uncertainty evaluation value, and the communication parameter stability coefficient matrix to obtain the communication abnormality type, and determining the fault diagnosis result according to the communication abnormality type.

[0132] The specific implementation methods of the above steps are described in detail below:

[0133] The specific implementation method of step S10 is: first, obtain the communication problem domain of the photovoltaic device. Secondly, determine the description language of the communication problem domain. Then, analyze the characteristics of the communication business domain based on the description language. Next, divide the boundaries of the communication business domain based on these characteristics. Finally, determine the boundary context of the communication business domain based on the boundary division results. Through the implementation of these sub-steps, the specific scope and background of the photovoltaic device communication problem can be clarified, laying the foundation for subsequent abnormal diagnosis.

[0134] The specific implementation of step S20 is as follows: first, multiple data acquisition channels are set, each channel corresponds to a type of communication real-time data. Then, corresponding types of communication real-time data are collected from photovoltaic devices through these data acquisition channels. Finally, the collected multi-type communication real-time data are classified and stored. Through multi-channel acquisition and classified storage, comprehensive communication real-time data samples can be obtained to provide basic data support for subsequent data analysis.

[0135] The specific implementation method of step S30 is: first, extreme point identification is performed based on real-time communication data. Secondly, the upper envelope and lower envelope are constructed for these extreme points using cubic spline interpolation. Then, the envelope mean of the upper and lower envelopes is calculated. Next, the envelope mean is subtracted from the real-time communication data to obtain the first component. Finally, the real-time communication data is subjected to intrinsic mode decomposition, in which the residual function is used as the stable component of the communication data, and the superposition of multiple intrinsic mode functions is used as the variable component of the communication data. Through these sub-steps, the real-time communication data can be effectively decomposed into stable components and variable components, laying the foundation for the subsequent extraction of communication anomaly features.

[0136] The specific implementation method of step S40 is: first, calculate the correlation terms between the stable components of communication data of different dimensions. Secondly, calculate the correlation terms between the variable components of communication data of different dimensions. Then, calculate the correlation terms between the gradient of the stable component of communication data and the gradient of the variable component of communication data. Next, introduce the time decay term and the error term. Finally, these correlation terms, time decay terms and error terms are weightedly combined to form a communication change relationship, and a communication parameter change coefficient matrix is ​​constructed based on this. In this way, a dynamic change model of communication parameters can be established to provide a basis for subsequent communication parameter stability analysis.

[0137] The specific implementation method of step S50 is: first, calculate the norm of the communication parameter variation coefficient matrix. Secondly, normalize the matrix. Then, introduce the time weight coefficient and the time scale parameter to construct the time decay function. Next, combine the normalized matrix with the time decay function to obtain the communication parameter stability coefficient matrix. After that, obtain the historical communication data of the photovoltaic device and pre-process it according to the preset data filtering rules. Finally, extract the characteristic parameters from the pre-processed historical data, and generate a historical characteristic sample set based on these characteristic parameters. Through the implementation of these sub-steps, the stability characteristics of the communication parameters can be quantitatively characterized, and the historical characteristic samples available for training can be obtained.

[0138] The specific implementation of step S60 is: first, obtain the preset feature weight coefficient, feature threshold upper limit, feature threshold lower limit and feature importance coefficient. Then, verify the validity of these preset parameters. These preset parameters play a key role in the subsequent feature extraction and optimization process, and their rationality and reliability need to be ensured.

[0139] The specific implementation method of step S70 is: first, input the feature weight coefficient and the communication parameter stability coefficient matrix into the feature mapping equation, which contains linear terms, nonlinear mapping terms and time derivative terms, and outputs the initial mapping value of the communication feature. Secondly, the initial mapping value and the upper and lower limits of the feature threshold are input into the feature constraint equation, which uses the sigmoid function for soft constraints and outputs the constrained eigenvalues. Then, the constrained eigenvalues ​​and the feature importance coefficients are input into the feature balance equation, which contains linear weight terms, nonlinear weight terms and acceleration weight terms, and outputs the balanced eigenvalues. Finally, the balanced eigenvalues ​​and the historical feature sample set are input into the feature optimization equation, which contains spatial regularization terms, time derivative terms, historical sample constraint terms and weighted integration terms, and outputs communication delay features, communication quality features and communication frequency features. Through the implementation of these sub-steps, the time-varying, nonlinear and historical dependence of communication features can be fully explored, and the accuracy and robustness of feature extraction can be improved.

[0140] The specific implementation method of step S80 is: first, construct a random deactivation neural network model. Then, train the model based on the historical feature sample set. Next, use the trained model to perform multiple forward propagation calculations. On this basis, calculate the abnormal probability values ​​and uncertainty evaluation values ​​of the communication delay characteristics, communication quality characteristics, and communication frequency characteristics respectively. The random deactivation mechanism can improve the generalization ability of the model, and these output indicators can comprehensively characterize the probability and uncertainty of communication anomalies.

[0141] The specific implementation method of step S90 is: first, the abnormal probability value, uncertainty assessment value and communication parameter stability coefficient matrix are used as cluster features. Secondly, cluster analysis is performed on these cluster features. Then, the type of communication anomaly is determined based on the cluster analysis results. Finally, the fault diagnosis result of the photovoltaic equipment is determined based on the type of communication anomaly. Through these sub-steps, multi-dimensional feature information can be comprehensively utilized to accurately identify and diagnose communication anomalies.

[0142] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed in a computer, they are used to execute the above-mentioned photovoltaic device communication abnormality diagnosis method.

[0143] A third aspect of the present invention provides a photovoltaic equipment communication anomaly diagnosis system, which includes the above-mentioned computer-readable storage medium.

[0144] Specifically, the principles of the present invention are: communication problem domain analysis, multi-dimensional communication data collection, communication data decomposition and modeling, communication feature extraction and optimization, and fault diagnosis based on anomaly detection. These steps are closely linked and together constitute a complete communication anomaly diagnosis process.

[0145] First, by analyzing the communication problem domain of photovoltaic equipment, the description language of communication problems is determined, and based on this, the boundaries of the communication business field are divided, which lays the foundation for subsequent data analysis and feature extraction. Secondly, multiple data acquisition channels are used to collect real-time communication data samples covering multiple dimensions such as delay, quality, and frequency from photovoltaic equipment. This multi-dimensional data acquisition can fully perceive the working status of the communication system and provide comprehensive data support for anomaly identification.

[0146] Then, the collected communication data is decomposed and calculated to divide it into stable components and variable components. Through advanced signal processing techniques such as intrinsic mode decomposition, the time-varying characteristics contained in the communication data can be effectively extracted. Next, based on the stable and variable components of the communication data, a communication parameter variation relationship model is established, and the communication parameter stability coefficient matrix is ​​further constructed. This method based on mathematical modeling can quantitatively characterize the dynamic change characteristics of communication parameters, laying the foundation for subsequent feature extraction and fault diagnosis.

[0147] On this basis, the present invention designs a series of feature extraction and optimization equations, which fully consider the complex characteristics of communication features such as time-varying, nonlinear and historical dependence. The feature mapping equation introduces nonlinear mapping and time derivative terms to enhance the ability of feature extraction; the feature constraint equation uses sigmoid function to implement soft constraints, avoiding the information loss caused by hard threshold truncation; the feature balance equation considers multiple factors such as linearity, nonlinearity and acceleration, and improves the dynamic balance of features; the feature optimization equation integrates multi-dimensional optimization strategies such as spatial regularization, time derivatives, and historical sample constraints, further improving the accuracy of features.

[0148] Finally, the present invention uses a random deactivation neural network model to evaluate the probability and uncertainty of communication anomalies based on the extracted communication features. Through cluster analysis, the abnormal features can be clustered into different fault types, thus providing a basis for subsequent fault diagnosis. Compared with the existing methods based on empirical rules or protocol analysis, this diagnostic method based on anomaly detection can more objectively and comprehensively identify communication faults, greatly improving the accuracy and reliability of diagnosis.

[0149] A specific embodiment 1 of the present invention is provided below. The specific implementation of each step in this embodiment 1 is described in detail as follows: The specific implementation of step S10 is as follows:

[0150] First, the communication problem domain Ω of the photovoltaic equipment is obtained. Second, the description language L of the communication problem domain is determined. Then, the characteristics of the communication business domain are analyzed based on the description language L. Next, based on these characteristics The communication business domain is divided into boundaries and the boundary context ΔΩ of the communication business domain is obtained.

[0151]

[0152] Among them, Ω represents the communication problem domain, L represents the description language of the problem domain, represents the characteristics of the communication business domain, and ΔΩ represents the boundary context of the communication business domain. Through the implementation of these sub-steps, the specific scope and background of the photovoltaic equipment communication problem can be clarified, laying the foundation for subsequent abnormal diagnosis.

[0153] The specific implementation of step S20 is as follows:

[0154] First, set up n data acquisition channels Each channel C i Corresponding to a type of communication real-time data Y i (t). Then, through these data acquisition channels {C i Collect corresponding types of communication real-time data from photovoltaic equipment Finally, the collected real-time data of multiple types of communication {Y i (t)} for classified storage.

[0155]

[0156] in, represents n data acquisition channels, Represents n types of real-time communication data. Through multi-channel acquisition and classified storage, comprehensive real-time communication data samples can be obtained to provide basic data support for subsequent data analysis.

[0157] The specific implementation of step S30 is as follows:

[0158] First, based on the real-time communication data Y(t), the extreme point is identified to obtain the extreme point set Secondly, the upper envelope e is constructed using the cubic spline interpolation method. max (t) and the lower envelope e min (t). Then, calculate the envelope mean Next, the envelope mean m(t) is subtracted from the real-time communication data Y(t) to obtain the first component h 1 (t) = Y(t) - m(t). Finally, h 1 (t) Perform eigenmode decomposition to obtain n eigenmode functions c i (t) and residual function r(t), where r(t) is the stable component of communication data S(t), As the communication data variation component V(t).

[0159]

[0160] Among them, Y(t) represents the real-time communication data, {x k} represents the set of extreme points, e max (t),e min (t) represents the upper and lower envelopes, m(t) represents the envelope mean, h 1 (t) represents the first component, {c i(t)} represents the intrinsic mode function, r(t) represents the residual function, S(t) represents the stable component of communication data, and V(t) represents the variable component of communication data. Through these sub-steps, the real-time communication data can be effectively decomposed into stable components and variable components, laying the foundation for the subsequent communication anomaly feature extraction.

[0161] The specific implementation of step S40 is as follows:

[0162] First, calculate the stable components S of communication data of different dimensions i (t),S j (t) 1 S i (t)S j (t). Secondly, calculate the communication data change components V of different dimensions i (t),V j (t) 2 V i (t)V j (t). Then, the gradient of the stable component of the communication data is calculated Gradient of the change component with communication data The relationship between Next, we introduce the time decay term Finally, these correlation terms, time decay terms and error terms are weighted together to form the communication change relationship M ij :

[0163]

[0164] Among them, S i (t),S j (t) represents the stable component of communication data in different dimensions, V i (t),V j (t) represents the communication data change components of different dimensions, α 1 ,α 2 ,α 3 ,α 4 is the weight coefficient, λ is the time attenuation coefficient, t 0 is the reference time, ∈ is the error term. Based on this, the communication parameter variation coefficient matrix M = {M ij}, a dynamic change model of communication parameters can be established to provide a basis for subsequent communication parameter stability analysis.

[0165] The specific implementation of step S50 is as follows:

[0166] First, the norm of the communication parameter variation coefficient matrix M is calculated. Secondly, M is normalized to obtain the normalized matrix Then, the time weight coefficient β and the time scale parameter μ are introduced to construct the time decay function Next, Combined with φ(t), we can get the communication parameter stability coefficient matrix K = {k ij},in Afterwards, obtain the historical communication data of the photovoltaic equipment According to the preset data filtering rules Preprocess it to get the historical feature sample set

[0167]

[0168] Among them, ∥M∥ represents the norm of matrix M, represents the normalized matrix, φ(t) represents the time decay function, K represents the communication parameter stability coefficient matrix, Y * Represents historical communication data, Indicates data filtering rules, Represents the historical feature sample set. Through these sub-steps, the stability characteristics of communication parameters can be quantitatively characterized, and historical feature samples for training can be obtained.

[0169] The specific implementation of step S60 is as follows:

[0170] First, obtain the preset feature weight coefficient matrix W = {w ij}. Secondly, obtain the preset feature threshold upper limit θ max and the feature threshold lower limit θ min Then, obtain the preset feature importance coefficient matrix P = {p ij},Q={q ij}, R = {r ij}. Finally, the validity of these preset parameters is verified to ensure their rationality and reliability. These preset parameters play a key role in the subsequent feature extraction and optimization process.

[0171] The specific implementation of step S70 is as follows:

[0172] First, the feature weight coefficient W and the communication parameter stability coefficient matrix K are input into the feature mapping equation The equation contains linear terms, nonlinear mapping terms φ(·) and time derivative terms, and outputs the initial mapping value F of the communication feature. Secondly, F and the upper and lower limits of the feature threshold θ max ,θ min Enter the feature constraint equation The equation uses the sigmoid function σ(·) for soft constraints and outputs the constrained eigenvalue G. Then, G and the feature importance coefficients P, Q, and R are input into the feature balance equation The equation contains linear weight terms, nonlinear weight terms and acceleration weight terms, and outputs the balanced eigenvalue H. Finally, H and the historical feature sample set Enter the characteristic optimization equation The equation contains spatial regularization term, time derivative term, historical sample constraint term and weighted integration term, and outputs the communication delay characteristic O 1 , Communication Quality Characteristics O 2 And communication frequency characteristics O 3 .

[0173]

[0174] Among them, F represents the initial mapping value of the communication feature, G represents the constrained eigenvalue, H represents the balanced eigenvalue, and O 1 ,O 2 ,O 3 They represent the communication delay characteristics, communication quality characteristics and communication frequency characteristics respectively, ∈ 1 ,∈ 2 ,∈ 3 ,∈ 4 Represents various error terms. Through these sub-steps, the time-varying, nonlinear and historical dependence of communication features can be fully explored to improve the accuracy and robustness of feature extraction.

[0175] The specific implementation of step S80 is as follows:

[0176] First, a random dropout neural network model is constructed. The forward propagation calculation of this model can be expressed as:

[0177] Z (l) =W (l) A (l-1) D (l) +b (l) ;

[0178] A (l) =f(Z (l) );

[0179] Among them, Z (l) is the weighted input of the lth layer, W (l) is the weight matrix, A (l-1) is the activation value of the previous layer, D (l) is the random dropout matrix, b (l) is the bias vector, and f is the activation function. Then, based on the historical feature sample set The model is trained. Next, the trained model is used to perform multiple forward propagation calculations. On this basis, the communication delay characteristics O are calculated respectively. 1 , Communication Quality Characteristics O 2 and communication frequency characteristics O 3 The abnormal probability value and uncertainty assessment The random dropout mechanism can improve the generalization ability of the model, and these output indicators can comprehensively characterize the probability and uncertainty of communication anomalies.

[0180] The specific implementation of step S90 is as follows:

[0181] First, the abnormal probability value Uncertainty assessment value and the communication parameter stability coefficient matrix K as clustering features Secondly, for these clustering features Perform cluster analysis and obtain cluster results Then, according to the cluster analysis results Determine the type of communication anomaly Finally, based on the communication exception type Determine the fault diagnosis results of photovoltaic equipment

[0182]

[0183] in, represents the clustering feature, represents the clustering result, Indicates the type of communication exception. Represents the fault diagnosis result. Through these sub-steps, multi-dimensional feature information can be comprehensively utilized to accurately identify and diagnose communication anomalies.

[0184] In order to better understand and implement the present invention, the following provides Example 2 of a specific application scenario of the present invention: According to the photovoltaic equipment communication abnormality diagnosis method proposed in the present invention, a photovoltaic power generation enterprise has set up a special communication diagnosis team to conduct a comprehensive diagnosis of the communication system of the photovoltaic power station. The specific implementation process is as follows:

[0185] 1. Communication Problem Domain Analysis

[0186] First, the diagnosis team conducted a detailed investigation of the communication system of the photovoltaic power station to understand its basic conditions such as network topology, communication protocol, and transmission medium. Through analysis, it was found that the communication system of the power station mainly had the following problem domains:

[0187] 1. Communication delay problem: Due to the long length of the optical fiber line and the presence of multiple relay stations in the middle, the communication delay time is long, which seriously affects the real-time performance of data transmission.

[0188] 2. Communication packet loss problem: In harsh natural environments, optical fiber lines are easily disturbed by factors such as lightning strikes and snow accumulation, resulting in a large number of data packets being lost, affecting the quality of communication.

[0189] 3. Unstable communication frequency problem: With the time-varying nature of light intensity, the acquisition frequency of each substation fluctuates greatly, which brings difficulties to the data processing and control decision-making of the host computer.

[0190] After further analysis, the diagnostic team determined that the description languages ​​of the three problem domains were: communication delay characteristics, communication quality characteristics, and communication frequency characteristics. Next, they divided the boundaries of the entire communication business field based on these characteristics and clarified the communication subsystems and data flows described by each characteristic.

[0191] Table 1 Communication problem domain and its description language

[0192] Problem Domain Description Language Corresponding subsystem Data Flow Communication delay Communication delay characteristics Access layer-aggregation layer-core layer Substation->Convergence station->Central station Communication packet loss Communication quality characteristics Access layer-aggregation layer Substation<->Convergence station Communication frequency Communication frequency characteristics Access Layer Substation->Convergence station

[0193] 2. Multi-dimensional Communication Data Collection

[0194] Based on the definition of the communication problem domain, the diagnosis team deployed multiple data acquisition channels at key locations of the photovoltaic power station to collect the operation data of the communication system in real time. Specifically, they include:

[0195] 1. Communication delay collection channel: deployed at several key nodes in the access layer, aggregation layer and core layer to collect end-to-end communication delay data.

[0196] 2. Communication quality collection channel: deployed on the key links of the access layer and the aggregation layer to collect indicators such as packet loss rate and bit error rate.

[0197] 3. Communication frequency collection channel: deployed on the access device of each substation to collect the data reporting frequency of each substation.

[0198] Through these multi-dimensional data collection channels, the diagnostic team accumulated a large amount of communication operation data in a short period of time, laying a solid foundation for subsequent fault diagnosis.

[0199] Table 2 Communication data acquisition channels and acquisition indicators

[0200] Acquisition channel Collection indicators Collection frequency Deployment location Communication delay End-to-end delay 1 second / time Access layer / aggregation layer / core layer Communication quality Packet loss rate, bit error rate 0.1 seconds / time Access layer / aggregation layer Communication frequency Data reporting frequency 0.5 seconds / time Access equipment for each substation

[0201] 3. Communication Data Decomposition and Modeling

[0202] With abundant communication operation data, the diagnosis team immediately began to analyze it in depth. First, they performed intrinsic mode decomposition on the collected communication data, dividing it into stable components and variable components. Taking the communication delay data Y(t) as an example, the specific decomposition process is as follows:

[0203] 1. Identify the extreme points of Y(t) and obtain the extreme point set

[0204] 2. Use cubic spline interpolation to construct the upper envelope e max (t) and the lower envelope e min (t).

[0205] 3. Calculate the envelope mean

[0206] 4. Subtract m(t) from Y(t) to get the first component h 1 (t) = Y(t) - m(t).

[0207] 5. For h 1 (t) Perform eigenmode decomposition to obtain n eigenmode functions c i (t) and residual function r(t).

[0208] Among them, r(t) is the stable component S(t) of the communication delay data, Then it is the variable component V(t). Similarly, the diagnosis team also performed similar decomposition processing on the communication quality and communication frequency data.

[0209] After obtaining the stable and variable components of the communication data, the diagnostic team set out to build a dynamic change model of the communication parameters. First, they calculated the correlation terms between the stable and variable components of different dimensions, such as α 1 S i (t)S j (t),α 2 V i (t)V j (t), And introduce the time decay term and error term∈, these terms are combined by weight to form the communication change relationship M ij :

[0210]

[0211] Among them, α 1 ,α 2 ,α 3 ,α 4 is the empirical weight coefficient, λ is the time attenuation coefficient, t 0Based on this communication change relationship, the diagnosis team constructed a communication parameter change coefficient matrix M = {M ij}.

[0212] Next, they calculated the norm of ∥M∥ and normalized M to obtain And introduce the time weight coefficient β and time scale parameter μ to construct the time decay function Finally, Combined with φ(t), we can get the communication parameter stability coefficient matrix K = {k ij},in

[0213] At the same time, the diagnostic team also obtained the historical communication data of the photovoltaic power station. And according to the preset data filtering rules It was preprocessed and a historical feature sample set was extracted from it.

[0214] Table 3 Communication data decomposition and modeling process

[0215]

[0216] 4. Communication Feature Extraction and Optimization

[0217] With the dynamic change model of communication parameters and historical feature samples, the diagnosis team began to extract and optimize communication anomaly features. First, they designed the following four feature extraction and optimization equations:

[0218] 1. Feature mapping equation:

[0219]

[0220] Where F is the initial mapping value of the communication feature, W is the feature weight coefficient matrix, φ(·) is the nonlinear mapping function, γ is the time derivative weight, ∈ 1 is the mapping error term.

[0221] 2. Characteristic constraint equation:

[0222]

[0223] Among them, G is the constrained eigenvalue, σ(·) is the sigmoid function, and θ max ,θ min is the upper and lower threshold of the feature,∈ 2 is the constrained error term.

[0224] 3. Characteristic balance equation:

[0225]

[0226] Among them, H is the balanced eigenvalue, P, Q, R are linear, nonlinear and acceleration weight matrices respectively, ∈ 3 is the balance error term.

[0227] 4. Feature optimization equation:

[0228]

[0229] Among them, O is the optimized eigenvalue, λ 1 ,λ 2 ,λ 3 ,λ 4 is the optimization coefficient, H * is the historical characteristic sample, ω i is the historical sample weight,∈ 4 is the optimization error term.

[0230] Through these feature extraction and optimization equations, the diagnosis team can fully explore the time-varying, nonlinear and historical dependence of communication features and improve the accuracy and robustness of feature expression. The final communication delay feature O 1 , Communication Quality Characteristics O 2 and communication frequency characteristics O 3 , which laid the foundation for subsequent fault diagnosis.

[0231] Table 4 Communication feature extraction and optimization process

[0232] step content Input / Output 1 Characteristic Mapping Equation <![CDATA[W,K,φ(·),γ,∈ 1 →F]]> 2 Characteristic Constraint Equation <![CDATA[F,θ max ,i min ,σ(·),∈ 2 →G]]> 3 Characteristic Balance Equation <![CDATA[P,Q,R,G,∈ 3 →H]]> 4 Characteristic Optimization Equation <![CDATA[H,λ 1 ,l 2 ,l 3 ,l 4 ,H * ,oh i ,∈ 4 →O 1 ,O 2 ,O 3 ]]>

[0233] 5. Fault diagnosis and analysis

[0234] After extracting the communication anomaly features, the diagnosis team then built a fault diagnosis model based on a random deactivation neural network. First, they built a random deactivation neural network in the following form:

[0235] Z (l) =W (l) A (l-1) D (l) +b (l) ;

[0236] A (l) =f(Z (l) );

[0237] Among them, Z (l) is the weighted input of the lth layer, W (l) is the weight matrix, A (l-1) is the activation value of the previous layer, D (l) is the random dropout matrix, b (l) is the bias vector and f is the activation function.

[0238] The diagnostic team then used a historical feature sample set The random dropout neural network was trained. After the training was completed, they used the model to analyze the communication delay characteristics. 1 , Communication Quality Characteristics O 2 and communication frequency characteristics O 3 After multiple forward propagation calculations, the abnormal probability values ​​of these three features were obtained respectively. and uncertainty estimates

[0239] Finally, the diagnosis team uses these abnormal probability values, uncertainty assessment values, and communication parameter stability coefficient matrix K as clustering features. Cluster analysis was performed. According to the clustering results They identified the main types of faults in the PV plant communication system And based on this, the specific fault diagnosis results can be determined

[0240] Table 5 Fault diagnosis and analysis process

[0241]

[0242] Through the above diagnostic analysis, the communication system of the photovoltaic power station mainly has the following types of faults:

[0243] 1. Communication delay is too high: Due to the long transmission delay between the access layer, aggregation layer and core layer, the end-to-end delay increases abnormally, affecting the real-time performance of the entire system.

[0244] 2. Serious communication packet loss failure: In harsh natural environments, the optical fiber links at the access layer and aggregation layer are disturbed, resulting in a large number of data packets being lost, which seriously affects the communication quality.

[0245] 3. Unstable communication frequency failure: With the fluctuation of light intensity, the data reporting frequency of each substation changes greatly, which brings difficulties to the data processing and control decision-making of the host computer.

[0246] In response to these diagnostic results, the company immediately developed targeted maintenance measures:

[0247] 1. Shorten the communication path length and reduce end-to-end delay by increasing the number of relay stations and optimizing network topology.

[0248] 2. Use optical fiber lines with stronger anti-interference capabilities, and strengthen monitoring and early warning of key links to reduce data loss.

[0249] 3. Adjust the control strategy of the substation collection frequency to better match it with the changes in lighting conditions and improve the stability of data reporting.

[0250] After the implementation of these maintenance measures, the operation status of the photovoltaic power station's communication system has been greatly improved. Key indicators such as communication delay, communication quality and communication frequency have reached normal levels, and the reliability of the entire photovoltaic power generation system has also been effectively improved.

[0251] It should be noted that the detailed explanation of the variables involved in the present invention is shown in Table 6.

[0252] Table 6 Variable explanation table

[0253]

[0254]

[0255] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for diagnosing abnormal communication of photovoltaic equipment, characterized in that: The following steps are involved: S10, obtaining a communication problem domain of the photovoltaic device, determining a description language of the communication problem domain, dividing the boundaries of the communication service domain based on the description language, and obtaining a boundary context of the communication service domain; S20, using multiple data acquisition channels to collect communication real-time data from the photovoltaic device, wherein one of the data acquisition channels corresponds to one type of the communication real-time data, and the multiple data acquisition channels are used to collect multiple types of the communication real-time data; S30, performing variation decomposition calculation on the real-time communication data to obtain a stable component of the communication data and a variation component of the communication data; S40, establishing a mapping function between the stable component of the communication data and the variable component of the communication data, recorded as a communication variable relationship, and establishing a communication parameter variable coefficient matrix according to the communication variable relationship; S50, calculating a communication parameter stability coefficient matrix based on the communication parameter variation coefficient matrix; Acquire historical communication data of the photovoltaic device, filter the historical communication data using data filtering rules, and generate a historical feature sample set; S60, obtaining a preset feature weight coefficient, a preset feature threshold upper limit, a preset feature threshold lower limit, and a preset feature importance coefficient; S70, optimizing and calculating the communication parameter stability coefficient matrix using a communication parameter optimization equation group to obtain communication delay characteristics, communication quality characteristics, and communication frequency characteristics; S80, training a random loss neural network model based on the historical feature sample set, and calculating abnormal probability values ​​and uncertainty assessment values ​​of the communication delay feature, the communication quality feature, and the communication frequency feature through multiple forward propagations; S90. Perform cluster analysis based on the abnormal probability value, the uncertainty evaluation value, and the communication parameter stability coefficient matrix to obtain a communication abnormality type, and determine a fault diagnosis result according to the communication abnormality type.

2. A photovoltaic equipment communication abnormality diagnosis method according to claim 1, characterized in that: The step S10 specifically includes: Step 101, obtaining a communication problem domain of a photovoltaic device; Step 102: Determine the description language of the communication problem domain; Step 103: Analyze the characteristics of the communication service field according to the description language; Step 104: dividing the communication service area into boundaries based on the characteristics; Step 105: determining the boundary context of the communication service domain according to the boundary division result; The step S20 specifically includes: Step 201, setting a plurality of data acquisition channels, each of the data acquisition channels corresponding to a type of communication real-time data; Step 202: collecting communication real-time data from the photovoltaic device through the multiple data collection channels; Step 203: Classify and store the real-time communication data collected by the multiple data collection channels.

3. A photovoltaic equipment communication abnormality diagnosis method according to claim 2, characterized in that: The step S30 specifically includes: Step 301: identifying extreme points based on real-time communication data; Step 302, constructing an upper envelope and a lower envelope for the extreme points using cubic spline interpolation; Step 303, calculating the envelope mean of the upper envelope and the lower envelope; Step 304: subtract the envelope mean value from the communication real-time data to obtain a first component; Step 305: Perform intrinsic mode decomposition on the communication real-time data to obtain multiple intrinsic mode functions and residual functions, wherein the residual function serves as a stable component of the communication data, and the superposition of the multiple intrinsic mode functions serves as a variable component of the communication data.

4. A photovoltaic equipment communication abnormality diagnosis method according to claim 3, characterized in that: The step S40 specifically includes: Step 401, calculating the correlation items between stable components of communication data of different dimensions; Step 402: Calculate the correlation items between the communication data change components of different dimensions; Step 403, calculating the correlation term between the gradient of the stable component of the communication data and the gradient of the variable component of the communication data; Step 404, introducing a time decay term and an error term; Step 405: Form a communication change relationship by weighted combination of the association term, the time decay term and the error term; Step 406: construct a communication parameter variation coefficient matrix based on the communication variation relationship.

5. A photovoltaic equipment communication abnormality diagnosis method according to claim 4, characterized in that: The step S50 specifically includes: Step 501, calculating the norm of the communication parameter variation coefficient matrix; Step 502: normalize the communication parameter variation coefficient matrix; Step 503: introducing a time weight coefficient and a time scale parameter; Step 504: construct a time decay function; Step 505: Combining the normalized matrix with the time decay function to obtain a communication parameter stability coefficient matrix; Step 506: Acquire historical communication data of the photovoltaic device; Step 507: pre-process the historical communication data according to preset data filtering rules; Step 508: extracting characteristic parameters from the pre-processed historical communication data; Step 509: Generate a historical feature sample set based on the feature parameters.

6. A photovoltaic equipment communication abnormality diagnosis method according to claim 5, characterized in that: The step S70 specifically includes: Step 701: Input the characteristic weight coefficient and the communication parameter stability coefficient matrix into the characteristic mapping equation, which contains linear terms, nonlinear mapping terms and time derivative terms, and output the initial mapping value of the communication characteristic; Step 702: input the communication feature initial mapping value and the feature threshold upper limit and feature threshold lower limit into a feature constraint equation, the feature constraint equation uses a sigmoid function to perform feature constraints, and outputs a constrained communication feature value; Step 703: input the constrained communication characteristic value and the characteristic importance coefficient into a characteristic balance equation, where the characteristic balance equation includes a linear weight term, a nonlinear weight term and an acceleration weight term, and output the balanced communication characteristic value; Step 704: input the balanced communication characteristic value and the historical characteristic sample set into a characteristic optimization equation, which contains a spatial regularization term, a time derivative term, a historical sample constraint term, and a weighted integration term, and outputs communication delay characteristics, communication quality characteristics, and communication frequency characteristics.

7. A photovoltaic equipment communication abnormality diagnosis method according to claim 6, characterized in that: The step S80 specifically includes: Step 801, constructing a random dropout neural network model; Step 802: training the random dropout neural network model based on the historical feature sample set; Step 803: Perform multiple forward propagation calculations using the random dropout neural network model; Step 804, respectively calculating the abnormal probability values ​​of the communication delay feature, the communication quality feature, and the communication frequency feature; Step 805: Calculate uncertainty evaluation values ​​of the communication delay characteristics, the communication quality characteristics, and the communication frequency characteristics respectively.

8. A photovoltaic equipment communication abnormality diagnosis method according to claim 7, characterized in that: The step S90 specifically includes: Step 901: taking the abnormal probability value, the uncertainty assessment value and the communication parameter stability coefficient matrix as clustering features; Step 902: performing cluster analysis on the cluster features; Step 903: Determine the type of communication anomaly according to the cluster analysis result; Step 904: Determine a fault diagnosis result of the photovoltaic device based on the communication abnormality type.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed in a computer, the method for diagnosing abnormal communication of a photovoltaic device according to any one of claims 1 to 8 is used to execute.

10. A photovoltaic equipment communication abnormality diagnosis system, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.

Images