A method, system and storage medium for evaluating the energy consumption of photovoltaic power generation equipment based on machine learning

Abstract

The present invention relates to the field of artificial intelligence technology, and discloses a method, a system and a storage medium for evaluating the energy consumption of photovoltaic power generation equipment based on machine learning, including obtaining the energy consumption data of the photovoltaic power generation equipment; using a neural network model based on the intelligent water drop algorithm to extract features from the energy consumption data of the photovoltaic power generation equipment to obtain the energy consumption characteristics of the photovoltaic power generation equipment; using an autoencoder to perform feature dimensionality reduction on the energy consumption characteristics of the photovoltaic power generation equipment to obtain the energy consumption dimensionality reduction characteristics of the photovoltaic power generation equipment; using a high-order neural network classification model to evaluate the energy consumption of the photovoltaic power generation equipment based on the energy consumption dimensionality reduction characteristics of the photovoltaic power generation equipment. The present invention improves the classification accuracy of the classifier under the influence of complex environments, thereby improving the accuracy of the energy consumption evaluation of photovoltaic power generation equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power generation equipment energy consumption assessment, and in particular to a photovoltaic power generation equipment energy consumption assessment method, system and storage medium based on machine learning. Background Art

[0002] As global energy demand continues to grow, photovoltaic power generation, as a clean and renewable energy source, is increasingly playing an increasingly important role in energy production. During the operation of photovoltaic power generation equipment, energy consumption assessment is crucial to optimizing power output, improving equipment efficiency, and reducing operating costs. However, since photovoltaic power generation equipment is usually exposed to a complex natural environment, the operating data of the equipment is affected by a variety of external factors such as light, temperature, and humidity, resulting in a high degree of complexity and uncertainty in its energy consumption data. In addition, problems such as scarce data samples and uneven data quality make it difficult for traditional energy consumption assessment methods to respond effectively, which in turn affects the accuracy and reliability of equipment energy efficiency optimization.

[0003] The Chinese invention patent with application number CN202011319075.8 proposes a method and device for evaluating the energy consumption of a centrifugal compressor, which belongs to the field of energy consumption evaluation. The method includes: determining the temperature variability index of the compressor in a preset calculation method, wherein the temperature variability index corresponds to the temperature of the current inlet state and the temperature of the outlet state of the compressor; determining the volume variability index of the compressor based on the temperature variability index; determining the operating power of the compressor according to the volume variability index, wherein the operating power represents the power status of the compressor under the current operating state; and determining the energy consumption evaluation result of the compressor according to the operating power. This evaluation method determines the volume variability index by the temperature variability index corresponding to the current compressor, and then determines the power status of the compressor under a certain operating state by the volume variability index. Because the value of the volume variability index is fixed under the same flow rate, the error of the energy consumption evaluation result can be reduced, so that the evaluation result can be closer to the actual result.

[0004] The Chinese invention patent with application number CN202210320596.8 proposes a method for evaluating the energy consumption of substation operation, including the following steps: step S1) determining the energy consumption composition of substation operation; step S2) collecting substation data; step S3) calculating the annual unit substation capacity energy consumption; step S4) constructing a substation operation energy consumption evaluation system architecture based on the analytic hierarchy process. The present invention uses the annual unit substation capacity energy consumption to quantitatively evaluate the substation operation energy consumption, and constructs a substation operation energy consumption evaluation system architecture based on the analytic hierarchy process to determine the substation operation energy consumption level, avoid the influence of subjective factors, and the evaluation results are more objective, intuitive and reliable, which can provide scientific guidance for the energy-saving optimization construction of substations.

[0005] The Chinese invention patent with application number CN201811580416.X proposes a vehicle energy consumption assessment method, device and storage medium. The method includes obtaining an energy consumption variable set of the vehicle to be tested, wherein the energy consumption variable set includes vehicle driving variables that characterize the vehicle driving conditions, vehicle state variables that characterize the use of vehicle components, and driving behavior variables that characterize the user's driving behavior; using the energy consumption variable set of the vehicle to be tested as the input of the energy consumption model to obtain the energy consumption of the vehicle to be tested, and the energy consumption model is pre-trained and established with the energy consumption variable set of the training vehicle and its corresponding energy consumption as samples. The method of the present application can specifically obtain the impact of each energy consumption variable on the energy consumption of the vehicle to be tested, thereby improving the pertinence and accuracy of the assessment.

[0006] The existing technology has the following problems that still need to be further solved:

[0007] 1. In the task of energy consumption assessment of photovoltaic power generation equipment, the data expansion method failed to effectively fill the data sparse areas, resulting in poor generalization ability of the model when there are insufficient samples.

[0008] 2. In the task of energy consumption assessment of photovoltaic power generation equipment, traditional feature extraction methods are inefficient when optimizing weights and parameters, and are difficult to adapt to data processing requirements in complex environments.

[0009] 3. In the task of energy consumption assessment of photovoltaic power generation equipment, traditional feature dimensionality reduction technology is difficult to fully capture the nonlinear structure in the data, resulting in insufficient feature expression ability after dimensionality reduction and large information loss.

[0010] 4. In the energy consumption assessment task of photovoltaic power generation equipment, traditional classifiers do not handle the uncertainty and fuzziness of energy consumption data well, resulting in low accuracy and reliability of energy consumption assessment classification. Summary of the invention

[0011] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for evaluating energy consumption of photovoltaic power generation equipment based on machine learning.

[0012] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0013] In a first aspect, a method for evaluating energy consumption of photovoltaic power generation equipment based on machine learning comprises the following steps:

[0014] Obtain energy consumption data of photovoltaic power generation equipment;

[0015] A neural network model based on the intelligent water drop algorithm is used to extract features from the energy consumption data of photovoltaic power generation equipment to obtain the energy consumption characteristics of photovoltaic power generation equipment;

[0016] The autoencoder is used to reduce the dimension of the energy consumption characteristics of photovoltaic power generation equipment to obtain the reduced dimension characteristics of the energy consumption of photovoltaic power generation equipment;

[0017] A high-order neural network classification model is used to evaluate the energy consumption of photovoltaic power generation equipment based on the dimensionality reduction features of the energy consumption of photovoltaic power generation equipment.

[0018] Furthermore, a neural network model based on the intelligent water drop algorithm is used to extract features from the energy consumption data of photovoltaic power generation equipment to obtain energy consumption features of photovoltaic power generation equipment, including:

[0019] Initialize the structure and parameters of the neural network;

[0020] At the beginning of each training iteration, a group of water droplets are generated by simulation, each of which carries a velocity vector and a position vector, and the velocity and direction of the water droplet are updated according to the gradient information on the path through path search in the feature space;

[0021] In the parameter space, the water droplets collide and merge according to the amount of information they carry and the path efficiency, and the parameters are updated during the optimization process;

[0022] Repeat the above steps until the preset stop iteration condition is met.

[0023] Furthermore, the activation function of the neural network is set to an adaptive noise suppression activation function, specifically:

[0024]

[0025] in, is the adaptive noise suppression activation function, is the energy consumption data of photovoltaic power generation equipment, is the parameter that controls the steepness of the function curve. is the parameter that determines the activation threshold.

[0026] Furthermore, in the parameter space, the collision and fusion are performed according to the amount of information carried by the water droplets and the path efficiency, and the parameter update in the optimization process is as follows:

[0027]

[0028]

[0029] in, For the The weights of the neural network at the iteration; For the The weights of the neural network at the iteration; is the learning rate of the neural network; is the gradient of the neural network's loss function with respect to the weights; For the The total energy of the water droplets in iterations; is the collection of water droplets participating in fusion; For the The bias of the neural network at this iteration; For the The bias of the neural network at this iteration; is the gradient of the loss function of the neural network with respect to the bias; For the The iteration The energy of a water drop.

[0030] Furthermore, an autoencoder is used to perform feature dimension reduction on the energy consumption characteristics of photovoltaic power generation equipment to obtain the reduced dimension features of the energy consumption of photovoltaic power generation equipment, including:

[0031] Initialize the weights and biases of the autoencoder;

[0032] Set up an adaptive feature recalibration layer to dynamically scale the response of each feature channel by dynamically adjusting the importance weights of each feature input to the autoencoder;

[0033] The adaptive feature recalibrated data is forward propagated through the encoder, and the encoder is used to convert the input data into a compressed low-dimensional feature vector through a multi-layer neural network;

[0034] Using quantum state interference mechanism on eigenvectors, by simulating superposition and interference of quantum phases;

[0035] The characteristic data after quantum state interference is reconstructed through a decoder;

[0036] Repeat the above steps until the preset stop iteration condition is met.

[0037] Furthermore, the quantum state interference mechanism is used on the eigenvector to simulate the superposition and interference of quantum phases, specifically:

[0038]

[0039] in, is the quantum state simulation function, is the output feature of the encoder, is the quantum phase parameter.

[0040] Furthermore, a high-order neural network classification model is used to evaluate the energy consumption of photovoltaic power generation equipment based on the dimensionality reduction features of the energy consumption of photovoltaic power generation equipment, including:

[0041] Initialize the weights and biases of high-order neural networks;

[0042] The data after the input feature dimension reduction is passed layer by layer through the high-order neural network;

[0043] Before the feature data reaches the decision-making layer, the fuzzy logic processing method is used to perform fuzzy classification on the feature data;

[0044] A nonlinearly scaled cross entropy loss function is used to calculate the difference between the actual output and the expected output;

[0045] Based on the loss function result, the gradient of each weight is calculated by the chain rule, and the Adam optimizer is used to update the weights and biases to minimize the loss function;

[0046] Repeat the above steps until the preset stop iteration condition is met.

[0047] Furthermore, before the feature data reaches the decision layer, the fuzzy logic processing method is used to perform fuzzy classification processing on the feature data, specifically:

[0048]

[0049] in, is the characteristic data obtained by fuzzy logic processing, is the parameter that controls the slope of the fuzzy logic function, For high-order neural networks The output features of the layer, is the threshold parameter of the fuzzy logic function.

[0050] Furthermore, after obtaining the energy consumption data of photovoltaic power generation equipment, the SMOTE method based on density variable weight is used to expand the energy consumption data of photovoltaic power generation equipment, including:

[0051] Calculate the local density for each photovoltaic power generation equipment sample point;

[0052] Calculate density weight according to the local density of each photovoltaic power generation equipment sample;

[0053] According to the density weight of each photovoltaic power generation equipment sample, a photovoltaic power generation equipment sample pair for synthesis is selected in the original data set;

[0054] For the selected photovoltaic power generation equipment sample pairs, new photovoltaic power generation equipment sample points are generated by vector interpolation method;

[0055] Repeat the above steps until the preset stop iteration condition is met.

[0056] Furthermore, for the selected photovoltaic power generation equipment sample pairs, new photovoltaic power generation equipment sample points are generated by vector interpolation method, specifically:

[0057]

[0058] in, For the newly synthesized photovoltaic power generation equipment sample point, is the characteristic deviation index, is the interpolation coefficient, is a nonlinear interpolation function, For the PV power generation equipment sample points, for of -The nearest neighbor The characteristic vector of the sample points of photovoltaic power generation equipment, To adjust the parameters of nonlinear effects, is the L2 norm.

[0059] In the second aspect, a photovoltaic power generation equipment energy consumption assessment system based on machine learning includes:

[0060] A data collection module, wherein the data collection module is used to obtain energy consumption data of photovoltaic power generation equipment;

[0061] A first data processing module, wherein the first data processing module uses a neural network model based on an intelligent water drop algorithm to extract features from energy consumption data of photovoltaic power generation equipment to obtain energy consumption features of photovoltaic power generation equipment;

[0062] A second data processing module, wherein the second data processing module uses an autoencoder to perform feature dimension reduction on the energy consumption characteristics of the photovoltaic power generation equipment to obtain the energy consumption dimension reduction characteristics of the photovoltaic power generation equipment;

[0063] A third data processing module, wherein the third data processing module uses a high-order neural network classification model to evaluate the energy consumption of photovoltaic power generation equipment based on the dimension reduction features of the energy consumption of photovoltaic power generation equipment;

[0064] An output module is used to output the energy consumption evaluation result of the photovoltaic power generation equipment.

[0065] In a third aspect, a computer-readable storage medium storing instructions is provided, wherein a computer program or instructions are stored in the storage medium. When the computer program or instructions are executed by an image processing device, a method for evaluating energy consumption of photovoltaic power generation equipment based on machine learning is implemented.

[0066] The present invention has the following beneficial effects:

[0067] 1. In the task of energy consumption assessment of photovoltaic power generation equipment, the present invention solves the problem of insufficient energy consumption data samples of photovoltaic power generation equipment through the SMOTE algorithm based on density variable weights. According to the local density difference, the proportion of data expansion is dynamically adjusted, and samples in low-density areas are expanded first to fill in data sparse areas and improve the effectiveness of data coverage.

[0068] 2. In the task of energy consumption assessment of photovoltaic power generation equipment, the present invention adopts the intelligent water drop algorithm to optimize the parameters of the neural network, and optimizes the weights and parameters of the neural network by simulating the path search and collision strategy of water droplets, thereby improving the efficiency and accuracy of feature extraction under complex data.

[0069] 3. In the task of energy consumption assessment of photovoltaic power generation equipment, the present invention adopts an autoencoder combined with a quantum state interference mechanism to train a feature dimensionality reduction model, effectively capturing the nonlinear structure in the data, reducing redundant information, and enhancing feature expression capabilities.

[0070] 4. In the task of energy consumption assessment of photovoltaic power generation equipment, the present invention adopts a high-order neural network classifier combined with a fuzzy logic processing module to process the uncertainty and fuzziness of energy consumption data, thereby improving the classification accuracy of the classifier under the influence of complex environments, thereby improving the accuracy of energy consumption assessment of photovoltaic power generation equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It is a flow chart of a method for energy consumption assessment of photovoltaic power generation equipment based on machine learning;

[0072] Figure 2 This is an experimental diagram for performance comparison of different sampling methods;

[0073] Figure 3 This is a performance comparison experiment diagram for different K values;

[0074] Figure 4 This is a comparison diagram of feature space distribution;

[0075] Figure 5 This is a comparison chart of the category balancing effects of different methods. DETAILED DESCRIPTION

[0076] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0077] like Figure 1 As shown, the embodiment of the present invention provides a method for evaluating energy consumption of photovoltaic power generation equipment based on machine learning, comprising the following steps S1 to S4:

[0078] S1. Obtain energy consumption data of photovoltaic power generation equipment;

[0079] In an optional embodiment of the present invention, the energy consumption data of the photovoltaic power generation equipment mainly comes from a real-time monitoring system, which continuously monitors the operating status of the equipment through sensors and automatically records data. The sensors can measure and provide various physical parameters of the equipment, such as voltage, current, temperature, etc. The collected data is stored in a structured JSON format to facilitate subsequent processing and analysis.

[0080] In one embodiment, the attributes of the data include:

[0081] is the equipment power (kW), is the voltage (V), is the current (A), is the ambient temperature (°C), is the light intensity (Lux), is the relative humidity (%), is the device status (0=off, 1=on), is the equipment efficiency (%), is the resistance (Ohms), is the capacitance (Farads).

[0082] In this embodiment, a piece of data of a photovoltaic power generation device at a specific time point may have the following form:

[0083] {

[0084] "timestamp": "2024-09-23T12:00:00Z", "device_id": "PV123456", "attributes": {" ":50," ":380," ":100," ":35," ":850," ":40," ":1," ":95," ":5," ":0.01}

[0085] }

[0086] It should be noted that this embodiment is only used to illustrate one data format and type of the present invention. In actual applications, the attributes of data are usually more than 10 attributes, and the number of attributes of data may reach dozens or even hundreds.

[0087] Furthermore, this embodiment labels the collected data by manual labeling. In one embodiment, the labeled categories include: low energy consumption, medium energy consumption and high energy consumption, a total of 3 categories.

[0088] It is understandable that in the task of the present invention, the collection, acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor model generalization ability and affect the accuracy of the model. The present invention uses the SMOTE (Synthetic Minority Over-sampling Technique) algorithm based on density variable weights to generate synthetic samples for energy consumption assessment of photovoltaic power generation equipment to make up for the lack of training samples and improve the generalization ability of machine learning models. On the basis of the traditional SMOTE algorithm, a non-uniform density weight adjustment strategy is adopted to automatically adjust the proportion of synthetic sample generation based on the local density difference of each sample point, thereby enhancing the target pertinence and efficiency of data expansion.

[0089] Specifically, the process of sample expansion based on the density-variable-weighted SMOTE algorithm is as follows:

[0090] S101. Density calculation: First, the local density of each photovoltaic power generation equipment sample point is calculated. Specifically, the k-nearest neighbor method is used to determine the number of other photovoltaic power generation equipment samples in the neighborhood of each photovoltaic power generation equipment sample point, which is used as a preliminary estimate of the density. Specifically, let The photovoltaic power generation equipment sample points are , calculate its - The average distance of the nearest neighbors to estimate the local density, expressed as:

[0091]

[0092] In the formula, express of -The nearest neighbor The characteristic vector of the photovoltaic power generation equipment samples, For the The local density of the PV power generation equipment samples, represents the L2 norm, is the pre-set neighborhood size, for and Preferably, Set to 10.

[0093] Furthermore, the associated sample weights give closer neighbors greater influence, making the density calculation more refined and sensitive to the local data structure. The calculation method is expressed as:

[0094]

[0095] In the formula, is a small constant that prevents division by zero. Preferably, Set to 0.01.

[0096] S102, perform density weight allocation, specifically calculate the density weight of each photovoltaic power generation equipment sample according to its local density, assign a lower weight to the area with higher density, and assign a higher weight to the area with lower density, to ensure that the artificially synthesized photovoltaic power generation equipment samples can effectively fill the data sparse area, set The local density of the photovoltaic power generation equipment samples is , the way to calculate its density weight is expressed as:

[0097]

[0098] In the formula, It is The density weight of the photovoltaic power generation equipment samples, is a parameter to adjust the density sensitivity, It is The local density of the PV power generation equipment samples; It is a hyperbolic tangent function, which uses nonlinear transformation to enhance the difference between high-density and low-density areas, so that the algorithm can focus more on low-density areas that really need data enhancement; is a parameter for adjusting the sensitivity of density conversion. Preferably, Set to 1.1, Set to 2.

[0099] S103, select and synthesize photovoltaic power generation equipment samples, select photovoltaic power generation equipment sample pairs in the original data set for synthesis. The selection process is adjusted based on the density weight, and photovoltaic power generation equipment sample points with high density weight are preferentially selected as the base points for synthesis. The method for calculating the probability of photovoltaic power generation equipment sample selection is expressed as:

[0100]

[0101] In the formula, Indicates selection The probability of a sample of photovoltaic power generation equipment, It is The density weight of the photovoltaic power generation equipment samples, It is The density weight of the photovoltaic power generation equipment samples, is the total number of PV power generation equipment samples before expansion.

[0102] S104. For the selected photovoltaic power generation equipment sample pair, a new photovoltaic power generation equipment sample point is generated by vector interpolation. The specific implementation of interpolation is to perform weighted average on the feature vectors of the two photovoltaic power generation equipment sample points. The weight is further adjusted by their density weight. Specifically, the selected photovoltaic power generation equipment sample pair is , the interpolation method for synthesizing new photovoltaic power generation equipment samples is expressed as:

[0103]

[0104] In the formula, It is a sample point of newly synthesized photovoltaic power generation equipment; From a uniform distribution The interpolation coefficients obtained by random sampling in , ensure that each new photovoltaic power generation equipment sample can be generated at a random position between the original photovoltaic power generation equipment samples; is a nonlinear interpolation function, is the characteristic deviation index.

[0105] Furthermore, the nonlinear interpolation function dynamically adjusts the interpolation intensity according to the actual distance between the photovoltaic power generation equipment samples, generates more complex and realistic data photovoltaic power generation equipment samples, and increases the diversity of photovoltaic power generation equipment sample generation. The calculation method is expressed as:

[0106]

[0107] In the formula, is the parameter that adjusts the nonlinear effect, is the distance between the selected photovoltaic power generation device sample pairs. Preferably, Set to 0.2.

[0108] Furthermore, the characteristic deviation index is used to optimize the quality of the synthesized photovoltaic power generation equipment samples according to the dynamic deviation correction synthesis mechanism. Specifically, it evaluates the deviation degree of the original data and dynamically adjusts it when the photovoltaic power generation equipment samples are synthesized to more truly reflect the essential characteristics of the data. The characteristic deviation index is calculated based on the skewness of the feature and is expressed as:

[0109]

[0110] In the formula, is the skewness of the feature, indicating The degree of skewness of the eigenvector is determined by The third-order central moment of is calculated.

[0111] S105, repeat the above process until a sufficient number of synthetic photovoltaic power generation equipment samples are generated or a preset data enhancement ratio is reached.

[0112] Furthermore, the generated synthetic photovoltaic power generation equipment samples are merged with the original data set to form a new training data set for subsequent model training.

[0113] This embodiment uses the SMOTE algorithm based on density variable weights to improve the pertinence of data expansion, improve the data quality in low-density areas, and enhance the generalization ability of the model.

[0114] S2. Using a neural network model based on the intelligent water drop algorithm to extract features from the energy consumption data of photovoltaic power generation equipment, and obtaining energy consumption features of photovoltaic power generation equipment;

[0115] In an optional embodiment of the present invention, in order to effectively capture and extract key information from the energy consumption data of photovoltaic power generation equipment, the present invention adopts a neural network structure based on the intelligent water droplet algorithm, and optimizes the weights and parameters in the feature extraction process by simulating the natural phenomenon of water droplets searching for the shortest path on different paths. A water droplet collision drive strategy is adopted on the basis of the traditional intelligent water droplet algorithm. This strategy enhances the dynamic optimization capability of feature extraction by simulating the collision and fusion process of multiple water droplets under interaction, and further improves the efficiency and accuracy of the algorithm in processing complex photovoltaic data.

[0116] Specifically, the training process of the neural network algorithm based on the intelligent water drop algorithm is as follows:

[0117] S201, initializing the structure and parameters of the neural network. In one embodiment, the initialization method is expressed as:

[0118]

[0119]

[0120] In the formula, It means that it obeys a specific distribution. Represents the neural network The weight matrix of the layer, Represents the neural network The bias vector of the layer, is the standard deviation of the neural network weight initialization, It means the mean is 0 and the variance is The normal distribution of Represents a normal distribution.

[0121] Furthermore, the activation function of the neural network is set to an adaptive noise suppression activation function. When processing photovoltaic power generation equipment data, these data often have a high degree of uncertainty and noise due to external environmental factors (such as weather changes). The adaptive noise suppression activation function has high robustness to abnormal data points and noise of photovoltaic power generation equipment, and has a higher suppression ability to noise and extreme values ​​by dynamically adjusting its response curve. Specifically, the calculation method of defining the adaptive noise suppression activation function is expressed as:

[0122]

[0123] In the formula, is the input of the adaptive noise suppression activation function, specifically the characteristic value of the photovoltaic power generation equipment data, such as the value 36 corresponding to the ambient temperature; is the adaptive noise suppression activation function, It is a parameter that controls the steepness of the function curve, and its value is dynamically adjusted according to the noise distribution of the input data; The parameters for determining the activation threshold are dynamically adjusted to match the central tendency of the data.

[0124] In each subsequent iteration, the values ​​of the activation function parameters are updated according to the following method to adapt to the changes in data:

[0125]

[0126]

[0127] In the formula, is the first learning rate parameter of the activation function, Is the second learning rate parameter of the activation function; is the mean value of the data feature, The standard deviation of the data feature; is the parameter that determines the activation threshold before updating, is the updated parameter for determining the activation threshold; is the parameter of the steepness of the control function curve before updating, is the parameter of the updated control function curve steepness.

[0128] S202. At the beginning of each training iteration, a group of water droplets are simulated and generated. Each water droplet carries a velocity vector and a position vector to simulate the search behavior in the feature space, and each water droplet represents a potential feature extraction path, that is, a set of parameters corresponding to the neural network. Furthermore, the water droplet searches for the path in the feature space, updates its speed and direction according to the gradient information on the path, and uses the water droplet collision drive strategy so that each water droplet can learn more information through interaction with other water droplets while exploring new paths. The update method of the velocity vector and position vector of the water droplet is expressed as:

[0129]

[0130]

[0131] In the formula, is the velocity vector of the water drop, is the position vector of the water drop, is the inertia coefficient of the water drop, is the learning rate of the droplet, is the gradient of the neural network's loss function with respect to the weight or bias, For the The velocity vector of the water droplet in the iteration, For the The velocity vector of the water droplet in the iteration, For the The position vector of the water droplet in the iteration, For the The position vector of the water droplet of the iteration. Preferably, Set to 1.3, It is set to 0.01, and the loss function of the neural network is calculated using cross entropy loss.

[0132] S203, the water droplets that meet in the parameter space will collide and merge according to the amount of information they carry and the efficiency of the path, and the parameters will be updated during the optimization process to enhance the balance between parameter exploration and utilization. Specifically, the collision and fusion process between water droplets is achieved by adjusting their paths, which can be expressed as:

[0133]

[0134]

[0135] in, For the The weights of the neural network at the iteration; For the The weights of the neural network at the iteration; is the learning rate of the neural network; is the gradient of the neural network's loss function with respect to the weights; For the The total energy of the water droplets in iterations; is the collection of water droplets participating in fusion; For the The bias of the neural network at this iteration; For the The bias of the neural network at this iteration; is the gradient of the loss function of the neural network with respect to the bias; For the The iteration The energy of a water droplet. Preferably, Set to 0.1.

[0136] Furthermore, the update of the weights during the collision process is calculated by considering the energy and collision efficiency of each droplet, expressed as:

[0137]

[0138] In the formula, is the path efficiency of the water droplet, For the The total energy of the water droplets in the iteration, is the change in efficiency due to the collision.

[0139] Furthermore, the calculation method of the efficiency change due to the collision is expressed as:

[0140]

[0141] in, represents the path efficiency before the collision, represents the path efficiency after the collision, and the path efficiency characterizes the entropy value of the set of water droplets involved in the fusion.

[0142] S204, repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.

[0143] This embodiment uses the intelligent water drop algorithm to improve the efficiency of the feature extraction process by optimizing weights and parameters, so that the model performs better when processing complex photovoltaic data.

[0144] S3, using an autoencoder to perform feature dimension reduction on the energy consumption characteristics of photovoltaic power generation equipment to obtain the energy consumption dimension reduction characteristics of photovoltaic power generation equipment;

[0145] In an optional embodiment of the present invention, the present invention adopts an autoencoder as a feature dimensionality reduction model, wherein the autoencoder is a neural network that learns data compression coding through unsupervised learning, and includes an encoder and a decoder: the encoder is responsible for compressing the input data into a low-dimensional feature representation, and the decoder attempts to reconstruct the original input data from the low-dimensional representation. In order to enhance the performance of the autoencoder and adapt to the characteristics of the energy consumption data of photovoltaic power generation equipment, the present invention adopts a coding mechanism based on quantum state interference on the basis of the traditional autoencoder, and considers the quantum phase information in the photovoltaic power generation equipment data after feature extraction, so as to better capture and compress the subtle patterns and characteristics in the data.

[0146] Specifically, the training process of the autoencoder algorithm is as follows:

[0147] S301, initialize the weight and bias of the autoencoder. In one embodiment, the weight and bias initialization formula of the autoencoder is:

[0148]

[0149]

[0150] In the formula, is the weight of the autoencoder, is the bias of the autoencoder; Represents the first The weight matrix of the layer; Represents the first The bias vector of the layer; is the standard deviation of the initialization weights of the autoencoder. Preferably, Set to 0.01.

[0151] S302, setting an adaptive feature recalibration layer, dynamically scaling the response of each feature channel by dynamically adjusting the importance weights of each feature input to the autoencoder, and defining the feature recalibration function of the adaptive feature recalibration layer as:

[0152]

[0153] In the formula, is the feature recalibration function; is the input feature of the adaptive feature recalibration layer, that is, the output feature of the feature extraction neural network, is the feature recalibration vector, each element of which corresponds to the recalibration factor of a feature in the input feature vector, calculated by the Sigmoid activation function; Represents element-wise multiplication.

[0154] S303, the data after adaptive feature recalibration is forward propagated through the encoder, and the encoder converts the input data into a compressed low-dimensional feature vector through a multi-layer neural network, which is expressed as:

[0155]

[0156]

[0157] In the formula, is the data feature after adaptive feature recalibration, The encoder The weighted input of the layer, The encoder The activation output of the layer.

[0158] Furthermore, the decoder partially reconstructs the output, and the calculation method is expressed as:

[0159]

[0160]

[0161] In the formula, The decoder is The weighted inputs of the layer; and are the weights and biases of the decoder; The decoder is The activation output of the layer; is the activation function of the autoencoder. Preferably, the activation function of the autoencoder adopts the Sigmoid activation function.

[0162] S304. Using the quantum state interference mechanism on the feature vector, by simulating the superposition and interference of quantum phases, the model's ability to capture the nonlinear structure of the input data is increased, while the ability to express the features is enhanced, expressed as:

[0163]

[0164] In the formula, is the output feature of the encoder, which is also the input feature of the decoder; is the quantum phase parameter; is the quantum state simulation function; is the output after quantum state interference. Preferably, .

[0165] S305, try to reconstruct the input data through the decoder. The structure of the decoder is symmetrical with that of the encoder. During the iterative training process, the calculation method of the loss function is expressed as:

[0166]

[0167] In the formula, is the number of samples input into the autoencoder for the current batch, The decoder outputs Reconstructed samples, The encoder input samples, is the loss function of the autoencoder.

[0168] Furthermore, according to the gradient descent method, the weights and biases of the autoencoder are updated using the loss function of the autoencoder, which is expressed as:

[0169]

[0170]

[0171] In the formula, Indicates a parameter update operation. is the learning rate of the autoencoder, is the symbol of partial derivative.

[0172] Furthermore, the learning rate of the autoencoder will adjust its value in each iteration according to the performance of the previous round of features, and the calculation method is expressed as:

[0173]

[0174] In the formula, Indicates The learning rate of the autoencoder for each round of iteration, Indicates The learning rate of the autoencoder for each round of iteration, The learning rate of the autoencoder affects the parameters, is the gradient of the loss function of the autoencoder with respect to the learning rate. Preferably, Set to 0.01.

[0175] S306, repeat the above steps until the preset stop iteration condition is met, which means the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.

[0176] This embodiment uses an autoencoder with the help of a quantum state interference mechanism to achieve effective dimensionality reduction of data, while retaining the expression of key features and improving the model's ability to capture nonlinear features.

[0177] S4. Use a high-order neural network classification model to evaluate the energy consumption of photovoltaic power generation equipment based on the dimensionality reduction features of the energy consumption of photovoltaic power generation equipment.

[0178] In an optional embodiment of the present invention, the dimensionality reduction features of the energy consumption of the photovoltaic power generation equipment are classified by a classifier, so as to obtain different classification categories, that is, corresponding to different energy consumption levels of the photovoltaic power generation equipment.

[0179] The present invention adopts a high-order neural network algorithm as a classifier model. In order to solve the problem of low data classification accuracy caused by the fact that the data of photovoltaic power generation equipment may be significantly affected by environmental factors (such as weather changes), on the basis of traditional high-order neural network algorithms, by integrating fuzzy logic principles into the decision-making layer of the neural network, it can better handle the uncertainty and ambiguity in the data.

[0180] Specifically, the training process of the high-order neural network classification algorithm is as follows:

[0181] S401, initializing the weights and biases of the high-order neural network. In one embodiment, the initialization method is expressed as:

[0182]

[0183]

[0184] In the formula, For high-order neural networks The weight of the layer, For high-order neural networks The bias of the layer, and are the number of neurons in the previous and current layers of the high-order neural network, respectively.

[0185] S402, the data after the input feature dimension reduction is transmitted layer by layer through the high-order neural network, and the process of data forward propagation is expressed as:

[0186]

[0187]

[0188] In the formula, For high-order neural networks The linear transformation result of the layer, For high-order neural networks The activation output of the layer, For high-order neural networks The activation output of the layer, is the Sigmoid activation function.

[0189] S403, before the features reach the decision layer, the fuzzy logic processing module is used to perform fuzzy classification processing on the features to improve the recognition ability of boundary conditions, which is expressed as:

[0190]

[0191] In the formula, represents the fuzzy logic function, is the feature obtained by fuzzy logic processing, For high-order neural networks The output features of the layer.

[0192] Furthermore, the fuzzy logic function adjusts the probability of each category according to the fuzzy set theory to deal with uncertainty and ambiguity. The calculation method is expressed as:

[0193]

[0194] In the formula, It is the parameter that controls the slope of the fuzzy logic function and determines the sensitivity of the function to the input; is the threshold parameter of the fuzzy logic function, which determines the midpoint of the activation function; and The adjustment of allows the model to handle the uncertainty and ambiguity of input features more flexibly. Preferably, Set to 0.3, Set to 0.2.

[0195] S404, using the cross entropy loss function to calculate the difference between the actual output and the expected output, and based on the cross entropy loss function, performing nonlinear scaling on the residual term in the loss function to enhance the optimization ability of the classifier when facing complex data, and adaptively adjusting the residual through the nonlinear function, so that the model is more sensitive to large errors during training and can more effectively reduce classification errors. The calculation method is expressed as:

[0196]

[0197] In the formula, is the loss function of the high-order neural network, is the total number of categories, It is the unique hot encoding of the true label corresponding to the samples of the current batch input to the high-order neural network; is the output of the fuzzy logic layer, indicating The predicted probability of the class; is the nonlinear weight of the loss function of the high-order neural network; is the nonlinear scaling factor of the loss function of the high-order neural network. Preferably, Set to 0.1, set to 0.01.

[0198] S405. According to the loss function result, the gradient of each weight is calculated by the chain rule, and the weight and bias are updated using the Adam optimizer to minimize the loss function, which is expressed as:

[0199]

[0200]

[0201] In the formula, For high-order neural networks The output features of the layer, For high-order neural networks The output features of the layer.

[0202] Furthermore, the Adam optimizer is used to update the weights and biases of the high-order neural network, expressed as:

[0203]

[0204]

[0205] In the formula, is the Adam optimizer function, is the learning rate of the high-order neural network; is a dynamically adjusted scaling factor that characterizes the importance of the input feature in the current layer. Preferably, Set to 0.01.

[0206] Furthermore, an adaptive weight adjustment strategy based on dynamic feature scaling is adopted. By adopting a dynamic feature scaling mechanism in each layer of the high-order neural network, adaptive adjustment is performed according to the weight contribution of the input feature, thereby improving the accuracy and generalization ability of the classification model. Specifically, the dynamically adjusted scaling coefficient is adaptively adjusted according to the back propagation error and feature gradient of the current layer, and the calculation method is expressed as:

[0207]

[0208] In the formula, is the feature scaling adjustment factor; is the reference scaling value, ensuring that the feature scaling factor is not adjusted when there is no error influence. Preferably, Set to 0.1.

[0209] S406, repeat the above steps until the preset stop iteration condition is met, which means the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations, preferably, the preset maximum number of iterations is set to 1000 times.

[0210] After the model training is completed, the trained model is used to evaluate the energy consumption of new samples of photovoltaic power generation equipment. In one embodiment, the collected raw data is input into the trained feature extraction and feature dimension reduction model for feature processing, and further, the processed features are input into the classifier model for classifier training to obtain the classification results. In this embodiment, the classification categories include: low energy consumption, medium energy consumption and high energy consumption, a total of 3 categories.

[0211] The high-order neural network in this embodiment is combined with a fuzzy logic processing module to effectively deal with the uncertainty and fuzziness in the data, making the energy consumption evaluation result of the classifier more accurate.

[0212] Experimental example

[0213] The experiment of density-variable-weighted SMOTE algorithm is to verify the effectiveness of density-variable-weighted SMOTE (Synthetic Minority Oversampling Technique) algorithm in photovoltaic equipment data enhancement. The comparison methods include original data (no oversampling), traditional SMOTE, ADASYN (Adaptive Synthetic Sampling Method), Borderline-SMOTE (Borderline Synthetic Minority Oversampling Technique) and this technology.

[0214] Figure 2 Experimental results show that this technology achieves the highest classification accuracy (86%±1%), which is 11% higher than the traditional SMOTE (75%±3%), and the standard deviation is significantly lower than other methods. The distribution of data points shows that the traditional method ignores local density features, and the accuracy fluctuates greatly in 5 repeated experiments (such as ADASYN fluctuates by 4%). However, this technology controls the fluctuation within 1% while maintaining high accuracy through dynamic density weight adjustment. Experimental results show that the weight allocation mechanism based on density variable weight can effectively identify data sparse areas and generate more representative synthetic samples, thereby improving the generalization ability of downstream classification models.

[0215] Figure 3In order to explore the sensitivity of the algorithm to the key parameter neighborhood size k, the experiment compared the performance changes of traditional SMOTE and this technology in the range of k=5 to 20. Traditional SMOTE reaches a peak accuracy of 75% when k=10, but when k>12, the accuracy drops rapidly due to the increase of noise samples (down to 70% when k=20), showing an obvious parabolic characteristic. In contrast, this technology maintains a stable high accuracy (84%-85%) in the range of k=8 to 16, and only slightly drops when k<6 due to density estimation distortion (83% when k=5), indicating that the hyperbolic tangent function in the dual adjustment mechanism of this technology suppresses the influence of extreme k values, while the nonlinear interpolation function compensates for the neighborhood division deviation through dynamic distance adjustment. Experiments show that the tolerance range of this technology for k values ​​is expanded by 60% (the effective range of the traditional method is k=8-12, and this paper is k=6-18), which significantly reduces the difficulty of parameter tuning.

[0216] Figure 4 It is indicated that through the principal component analysis (PCA) dimension reduction visualization, this experiment compares the distribution characteristics of the original data, traditional SMOTE and samples generated by this technology. The original data presents a multi-cluster aggregation form in the principal component space (contour coefficient 0.62), and the samples generated by traditional SMOTE present a linear band distribution along the feature axis (contour coefficient 0.51), and a large number of outliers are generated at the category boundary (accounting for 12%). The samples generated by this technology better maintain the original cluster structure (contour coefficient 0.59), and the nonlinear interpolation function φ makes the new samples present a reasonable scattering distribution (the maximum local density difference is reduced by 37%). In low-density areas (such as coordinates [-2.5,1.8]), the number of samples generated by this technology is 2.3 times higher than that of traditional SMOTE, while the number of samples in high-density areas (such as coordinates [1.2,-0.5]) is reduced by 45%, which verifies that the density weight allocation mechanism can accurately locate the area to be enhanced and avoid the feature space distortion problem caused by oversampling.

[0217] Figure 5 It is stated that in order to address the category imbalance problem in photovoltaic equipment data (the original data category ratio is 3:2:1), this experiment uses F1-score to evaluate the balancing effects of different methods. The original data has an F1-score of only 65% ​​in the minority class (category 3), and traditional SMOTE increases it to 70%, but the performance of the majority class (category 1) decreases by 3%. This technology achieves balanced improvement in the three categories (category 1: 82%, category 2: 84%, category 3: 81%), of which the minority class has increased by 16%. The number of samples generated by this technology in the low-density area of ​​category 3 is 2.8 times that of traditional SMOTE, indicating that the selective enhancement strategy driven by density weight can break through the limitations of simple linear interpolation and achieve adaptive category balance.

[0218] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0219] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0220] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0221] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0222] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for evaluating energy consumption of photovoltaic power generation equipment based on machine learning, characterized in that: The following steps are involved: Obtain energy consumption data of photovoltaic power generation equipment; A neural network model based on the intelligent water drop algorithm is used to extract features from the energy consumption data of photovoltaic power generation equipment to obtain the energy consumption characteristics of photovoltaic power generation equipment, including: Initialize the structure and parameters of the neural network; set the activation function of the neural network to the adaptive noise suppression activation function, specifically: in, is the adaptive noise suppression activation function, is the energy consumption data of photovoltaic power generation equipment, is the parameter that controls the steepness of the function curve. is the parameter that determines the activation threshold; At the beginning of each training iteration, a group of water droplets are generated by simulation, each of which carries a velocity vector and a position vector, and the velocity and direction of the water droplet are updated according to the gradient information on the path through path search in the feature space; In the parameter space, the water droplets collide and merge according to the amount of information carried by the water droplets and the path efficiency, and the parameters are updated during the optimization process; wherein the parameter update during the optimization process is specifically as follows: , in, For the The weights of the neural network at the iteration; For the The weights of the neural network at the iteration; is the learning rate of the neural network; is the gradient of the neural network's loss function with respect to the weights; For the The total energy of the water droplets in iterations; is the collection of water droplets participating in fusion; For the The bias of the neural network at this iteration; For the The bias of the neural network at this iteration; is the gradient of the loss function of the neural network with respect to the bias; For the The iteration The energy of a water drop; The updating method of the velocity vector and position vector of the water droplet is expressed as: , In the formula, is the velocity vector of the water drop, is the position vector of the water drop, is the inertia coefficient of the water drop, is the learning rate of the droplet, is the gradient of the neural network's loss function with respect to the weight or bias, For the The velocity vector of the water droplet in the iteration, For the The velocity vector of the water droplet in the iteration, For the The position vector of the water droplet in the iteration, For the The position vector of the water droplet in the iteration; The update of the weights in the collision process is calculated by considering the energy and collision efficiency of each water droplet, expressed as: In the formula, is the path efficiency of the water droplet, For the The total energy of the water droplets in the iteration, is the amount of change in efficiency due to the collision; Repeat the above steps until the preset stop iteration condition is met; The autoencoder is used to reduce the dimension of the energy consumption characteristics of photovoltaic power generation equipment to obtain the reduced dimension characteristics of the energy consumption of photovoltaic power generation equipment; A high-order neural network classification model is used to evaluate the energy consumption of photovoltaic power generation equipment based on the dimensionality reduction features of the energy consumption of photovoltaic power generation equipment.

2. A photovoltaic power generation equipment energy consumption assessment method based on machine learning according to claim 1, characterized in that: A high-order neural network classification model is used to evaluate the energy consumption of photovoltaic power generation equipment based on the dimensionality reduction features of the energy consumption of photovoltaic power generation equipment, including: Initialize the weights and biases of high-order neural networks; The data after the input feature dimension reduction is passed layer by layer through the high-order neural network; Before the feature data reaches the decision-making layer, the fuzzy logic processing method is used to perform fuzzy classification on the feature data; A nonlinearly scaled cross entropy loss function is used to calculate the difference between the actual output and the expected output; Based on the loss function result, the gradient of each weight is calculated by the chain rule, and the Adam optimizer is used to update the weights and biases to minimize the loss function; Repeat the above steps until the preset stop iteration condition is met.

3. A photovoltaic power generation equipment energy consumption assessment method based on machine learning according to claim 2, characterized in that: Before the feature data reaches the decision-making layer, the fuzzy logic processing method is used to perform fuzzy classification on the feature data, specifically: in, is the characteristic data obtained by fuzzy logic processing, is the parameter that controls the slope of the fuzzy logic function, For high-order neural networks The output features of the layer, is the threshold parameter of the fuzzy logic function.

4. The method for evaluating energy consumption of photovoltaic power generation equipment based on machine learning according to claim 1, characterized in that: After obtaining the energy consumption data of photovoltaic power generation equipment, the SMOTE method based on density variable weight is used to expand the energy consumption data of photovoltaic power generation equipment, including: Calculate the local density for each photovoltaic power generation equipment sample point; Calculate density weight according to the local density of each photovoltaic power generation equipment sample; According to the density weight of each photovoltaic power generation equipment sample, a photovoltaic power generation equipment sample pair for synthesis is selected in the original data set; For the selected photovoltaic power generation equipment sample pairs, new photovoltaic power generation equipment sample points are generated by vector interpolation method; Repeat the above steps until the preset stop iteration condition is met.

5. The method for evaluating energy consumption of photovoltaic power generation equipment based on machine learning according to claim 4, characterized in that: For the selected photovoltaic power generation equipment sample pairs, new photovoltaic power generation equipment sample points are generated by vector interpolation method, specifically: in, For the newly synthesized photovoltaic power generation equipment sample point, is the characteristic deviation index, is the interpolation coefficient, is a nonlinear interpolation function, For the PV power generation equipment sample points, for of -The nearest neighbor The characteristic vector of the sample points of photovoltaic power generation equipment, To adjust the parameters of nonlinear effects, is the L2 norm.

6. A photovoltaic power generation equipment energy consumption assessment system based on machine learning, comprising implementing the method as claimed in any one of claims 1 to 5, characterized in that: A data collection module, wherein the data collection module is used to obtain energy consumption data of photovoltaic power generation equipment; A first data processing module, wherein the first data processing module uses a neural network model based on an intelligent water drop algorithm to extract features from energy consumption data of photovoltaic power generation equipment to obtain energy consumption features of photovoltaic power generation equipment; A second data processing module, wherein the second data processing module uses an autoencoder to perform feature dimension reduction on the energy consumption characteristics of the photovoltaic power generation equipment to obtain the energy consumption dimension reduction characteristics of the photovoltaic power generation equipment; A third data processing module, wherein the third data processing module uses a high-order neural network classification model to evaluate the energy consumption of photovoltaic power generation equipment based on the dimension reduction features of the energy consumption of photovoltaic power generation equipment; An output module is used to output the energy consumption evaluation result of the photovoltaic power generation equipment.

7. A computer-readable storage medium storing instructions, characterized in that: The storage medium stores a computer program or instruction, and when the computer program or instruction is executed, the method as claimed in any one of claims 1 to 5 is implemented.

Images