Cleaning equipment energy efficiency evaluation and optimization method and device based on artificial intelligence

Abstract

The invention provides a cleaning equipment energy efficiency evaluation and optimization method and device based on artificial intelligence, and the method comprises the steps: obtaining data based on log records in the operation process of cleaning equipment, and marking the data; performing data expansion on the data based on the quantum entangled state generative adversarial network to obtain expanded data; training a feature extraction model based on simulated load balance based on the expanded data; training a feature dimension reduction model based on the data after feature extraction; training a classifier model based on the dimensionality-reduced data; and evaluating the energy efficiency of the cleaning equipment based on the trained feature extraction model, the feature dimension reduction model and the classifier model. In the mode, the inherent relevance and expression ability of data are improved by simulating the characteristics of quantum entanglement, the weight update in each training period is adaptively adjusted by using a dynamic variable weight strategy, the structure of a potential space is optimized by simulating the distribution of input data, and the weight sharing level is dynamically adjusted according to the characteristics of the input data.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an artificial intelligence-based cleaning equipment energy efficiency evaluation and optimization method and device. Background Art

[0002] Energy efficiency evaluation and optimization of cleaning equipment usually covers multiple aspects of cleaning equipment operation, including equipment performance monitoring, energy consumption data analysis, operating parameter adjustment and optimization strategy design, with the goal of achieving efficient operation of equipment and reducing energy consumption. In practical applications, energy efficiency evaluation and optimization of cleaning equipment faces the challenge of how to accurately evaluate equipment energy efficiency and improve overall efficiency, especially in data collection, processing and intelligent analysis.

[0003] At present, existing evaluation methods often rely on traditional data processing technologies, which have obvious limitations when facing large-scale, high-dimensional and complex distributed data generated during the operation of cleaning equipment, and are difficult to adapt to the diverse needs of cleaning tasks and dynamic changes under different working conditions. Therefore, developing an efficient and accurate energy efficiency evaluation and optimization method based on artificial intelligence has become an important demand in current research and industrial applications. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide an artificial intelligence-based cleaning equipment energy efficiency evaluation and optimization method and device, so as to simulate the characteristics of quantum entanglement to enhance the intrinsic correlation and expression ability of data, and solve the problem of insufficient sample diversity in traditional generative adversarial networks; utilize a dynamic variable weight strategy to adaptively adjust the weight update in each training cycle, thereby solving the problems of gradient vanishing, gradient explosion or falling into local optimal solutions that traditional neural networks may encounter in feature extraction; optimize the structure of the latent space by simulating the distribution of input data, thereby solving the problem that traditional feature dimensionality reduction technology may ignore the intrinsic structure and distribution of data; dynamically adjust the weight sharing level according to the characteristics of the input data, thereby solving the problems of low efficiency and poor model generalization ability of traditional classifiers when processing different data distributions.

[0005] In a first aspect, an embodiment of the present invention provides an artificial intelligence-based cleaning equipment energy efficiency evaluation and optimization method, the method comprising: acquiring data based on log records during the operation of the cleaning equipment, and labeling the data; wherein the labeled categories are used to characterize the energy efficiency of the cleaning equipment; performing data expansion on the data based on a generative adversarial network of quantum entangled states to obtain expanded data; training a feature extraction model based on simulated load balancing based on the expanded data, and the feature extraction model outputs the feature extracted data; wherein the weight update in each training cycle of the feature extraction model is adaptively adjusted through a dynamic variable weight strategy; training a feature dimension reduction model based on the feature extracted data, and the feature dimension reduction model outputs the reduced dimension data; wherein the feature dimension reduction model is determined based on an autoencoder algorithm of a Gaussian process latent variable; training a classifier model based on the reduced dimension data; wherein a deep neural decision tree algorithm based on adaptive weight sharing is used as the classification algorithm of the classifier model; and evaluating the energy efficiency of the cleaning equipment based on the trained feature extraction model, feature dimension reduction model and classifier model.

[0006] In an optional embodiment of the present application, the above-mentioned generative adversarial network of quantum entangled state includes a generator and a discriminator, and the input layer of the generator adopts the representation form of quantum state.

[0007] In an optional embodiment of the present application, the above-mentioned generative adversarial network trains the generator and the discriminator through the following steps: initializing the generative adversarial network; in each iteration, the generator receives the quantum state representation of random noise and target data, simulates the generation of entangled states through quantum gate operations, and outputs data samples; the discriminator obtains real samples and data samples output by the generator, and updates the weights of the discriminator based on the posterior probabilities of the real samples and the data samples output by the generator; back propagates based on the gradient calculated based on a preset loss function, and updates the parameters and quantum state parameters of the generative adversarial network; determines whether the preset stop iteration condition is met; if the stop iteration condition is met, stops the iteration and determines the final generative adversarial network; if the stop iteration condition is not met, continues to iterate.

[0008] In an optional embodiment of the present application, the above-mentioned step of training a feature extraction model based on simulated load balancing based on expanded data includes: initializing the weights and biases of the network layer of the feature extraction model; at the beginning of each training batch, dynamically adjusting the weights of the network layer of the feature extraction model based on the simulated load balancing according to the training results of the previous batch; when the data stream passes through the network of the feature extraction model for feature extraction, based on a preset load balancing optimization algorithm, dynamically adjusting the training speed and accuracy of the feature extraction model using the response time and energy consumption of the current network; adjusting the network structure of the feature extraction model based on the features extracted by the feature extraction model, and correcting the network parameters of the feature extraction model based on the contribution weight; judging whether the preset stop iteration condition is met; if the stop iteration condition is met, stopping the iteration and determining the final feature extraction model; if the stop iteration condition is not met, continuing the iteration.

[0009] In an optional embodiment of the present application, the above-mentioned step of training a feature dimensionality reduction model based on the data after feature extraction includes: initializing the parameters of the autoencoder and the Gaussian process; the encoder transforms the data after feature extraction to obtain a representation of the latent space; the Gaussian process models the distribution of the latent variables; sparse matrix decomposition is used in the latent space to reduce the data dimension; back propagation is performed according to the loss function of the autoencoder to adjust the parameters of the feature dimensionality reduction model; it is determined whether a preset stop iteration condition is met; if the stop iteration condition is met, the iteration is stopped and the final feature dimensionality reduction model is determined; if the stop iteration condition is not met, the iteration is continued.

[0010] In an optional embodiment of the present application, the above-mentioned step of training a classifier model based on the reduced-dimensional data includes: initializing the structure and node parameters of a deep neural decision tree; wherein the parameters of each deep neural decision tree node include decision weights and biases; outputting the output value of each node through a forward propagation process; dynamically adjusting the weight sharing strategy according to the flow of samples in the decision tree; evaluating the complexity and performance of the classifier model during training, and dynamically adjusting the depth and number of nodes of the decision tree; determining whether a preset stop iteration condition is met; if the stop iteration condition is met, stopping the iteration and determining the final classifier model; if the stop iteration condition is not met, continuing the iteration.

[0011] In an optional embodiment of the present application, the above-mentioned step of outputting the output value of each node through the forward propagation process includes: calculating the decision result of each input sample at each node according to the current weight and bias for each input sample, and determining the path of each input sample in the decision tree.

[0012] In an optional embodiment of the present application, the above-mentioned step of evaluating the energy efficiency of the cleaning equipment based on the trained feature extraction model, feature dimension reduction model and classifier model includes: obtaining input data of the cleaning equipment; inputting the input data into the trained feature extraction model, and outputting the input data after feature extraction; inputting the input data after feature extraction into the trained feature dimension reduction model, and outputting the input data after dimension reduction; inputting the input data after dimension reduction into the trained classifier model, and outputting the energy efficiency of the cleaning equipment.

[0013] In an optional embodiment of the present application, after the above-mentioned step of evaluating the energy efficiency of the cleaning equipment based on the trained feature extraction model, feature dimensionality reduction model and classifier model, the method also includes: determining operational optimization suggestions for the cleaning equipment based on the energy efficiency of the cleaning equipment; wherein the operational optimization suggestions include at least one of the following: adjusting the operating parameters of the cleaning equipment, optimizing the cleaning agent usage ratio or improving the equipment operation logic.

[0014] In the second aspect, the embodiment of the present invention also provides an energy efficiency evaluation and optimization device for cleaning equipment based on artificial intelligence, the device comprising: a data acquisition and annotation module, used to acquire data based on log records during the operation of the cleaning equipment, and annotate the data; wherein the annotated category is used to characterize the energy efficiency of the cleaning equipment; a data expansion module, used to expand the data based on the generative adversarial network of quantum entangled states to obtain expanded data; a feature extraction model training module, used to train a feature extraction model based on simulated load balancing based on the expanded data, and the feature extraction model outputs the feature extracted data; wherein, through the dynamic variable weight strategy, The weight update in each training cycle of the feature extraction model is adaptively adjusted; a feature dimension reduction model training module is used to train the feature dimension reduction model based on the data after feature extraction, and the feature dimension reduction model outputs the reduced data; wherein the feature dimension reduction model is determined based on the autoencoder algorithm of the Gaussian process latent variable; a classifier model training module is used to train the classifier model based on the reduced data; wherein a deep neural decision tree algorithm based on adaptive weight sharing is used as the classification algorithm of the classifier model; a cleaning equipment energy efficiency evaluation module is used to evaluate the energy efficiency of the cleaning equipment based on the trained feature extraction model, feature dimension reduction model and classifier model.

[0015] The embodiments of the present invention bring the following beneficial effects: The embodiment of the present invention provides a method and device for energy efficiency evaluation and optimization of cleaning equipment based on artificial intelligence. Data expansion is performed through a generative adversarial network based on quantum entangled states, and the characteristics of quantum entanglement are simulated to enhance the intrinsic correlation and expression ability of data, thereby solving the problem of insufficient sample diversity in traditional generative adversarial networks. A fully connected neural network with simulated load balancing optimization is adopted as a feature extraction model, and a dynamic variable weight strategy is utilized to adaptively adjust the weight update in each training cycle, thereby solving the problems of gradient vanishing, gradient exploding, or falling into local optimal solutions that may be encountered by traditional neural networks in feature extraction. An autoencoder based on Gaussian process latent variables is adopted to realize feature dimensionality reduction, and the structure of the latent space is optimized by simulating the distribution of input data, thereby solving the problem that traditional feature dimensionality reduction technology may ignore the intrinsic structure and distribution of data. A deep neural decision tree based on adaptive weight sharing is adopted as a classifier, and the weight sharing level is dynamically adjusted according to the characteristics of the input data, thereby solving the problems of low efficiency and poor model generalization ability of traditional classifiers when processing different data distributions.

[0016] Other features and advantages of the present disclosure will be set forth in the following description, or some features and advantages may be inferred or unambiguously determined from the description, or may be learned by implementing the above-mentioned technology of the present disclosure.

[0017] In order to make the above-mentioned objectives, features and advantages of the present disclosure more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A flowchart of a cleaning equipment energy efficiency evaluation and optimization method based on artificial intelligence provided by an embodiment of the present invention; Figure 2 A schematic diagram of the structure of a cleaning equipment energy efficiency evaluation and optimization device based on artificial intelligence provided by an embodiment of the present invention; Figure 3 A schematic diagram of the structure of another cleaning equipment energy efficiency evaluation and optimization device based on artificial intelligence provided by an embodiment of the present invention; Figure 4 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] At present, the development of an efficient and accurate energy efficiency evaluation and optimization method based on artificial intelligence has become an important demand in current research and industrial applications. By combining artificial intelligence technology, it is possible to more comprehensively analyze the operating status of equipment, predict potential energy efficiency bottlenecks, and achieve the economic, safety and environmental protection goals of cleaning equipment, providing support for the green transformation of the industry.

[0022] However, the existing cleaning equipment energy efficiency evaluation methods have the following defects: 1. In existing generative adversarial network applications, common challenges include insufficient sample diversity, which limits the generalization ability of the model in practical applications.

[0023] 2. Traditional neural networks often encounter the problem of gradient disappearance or explosion during the feature extraction stage, which affects the stability of model training and the efficiency of feature extraction.

[0024] 3. Existing feature dimensionality reduction techniques often ignore the intrinsic structure and distribution of the data, resulting in the inability of the reduced features to fully express the important information of the original data, affecting the interpretability and prediction accuracy of the model.

[0025] 4. In classifier design, there is a lack of effective mechanisms to adapt to different data distributions, resulting in poor efficiency and accuracy of the model when processing diverse data.

[0026] Based on this, an embodiment of the present invention provides an artificial intelligence-based cleaning equipment energy efficiency evaluation and optimization method and device, which performs data expansion through a generative adversarial network based on quantum entangled states, simulates the characteristics of quantum entanglement to enhance the intrinsic correlation and expression ability of data, and solves the problem of insufficient sample diversity in traditional generative adversarial networks; adopts a fully connected neural network with simulated load balancing optimization as a feature extraction model, and uses a dynamic variable weight strategy to adaptively adjust the weight update in each training cycle, thereby solving the problems of gradient vanishing, gradient explosion, or falling into local optimal solutions that may be encountered by traditional neural networks in feature extraction; adopts an autoencoder based on Gaussian process latent variables to achieve feature dimensionality reduction, and optimizes the structure of the latent space by simulating the distribution of input data, thereby solving the problem that traditional feature dimensionality reduction technology may ignore the intrinsic structure and distribution of data; adopts a deep neural decision tree based on adaptive weight sharing as a classifier, and dynamically adjusts the weight sharing level according to the characteristics of the input data, thereby solving the problems of low efficiency and poor model generalization ability of traditional classifiers when processing different data distributions.

[0027] To facilitate understanding of this embodiment, an artificial intelligence-based cleaning equipment energy efficiency evaluation and optimization method disclosed in an embodiment of the present invention is first introduced in detail.

[0028] The embodiment of the present invention provides a cleaning equipment energy efficiency evaluation and optimization method based on artificial intelligence, see Figure 1 A flowchart of a cleaning equipment energy efficiency evaluation and optimization method based on artificial intelligence is shown, and the cleaning equipment energy efficiency evaluation and optimization method based on artificial intelligence includes the following steps: Step S102, acquiring data based on log records during the operation of the cleaning device, and labeling the data; wherein the labeled category is used to characterize the energy efficiency of the cleaning device.

[0029] The data collection of the present invention is derived from the log records during the operation of the cleaning equipment. The data storage format can adopt a structured JSON (JavaScript Object Notation) format. In one embodiment, the attributes of the data may include: Ra1 is the total energy consumption of the equipment, in kilowatt-hour (kWh); Ra2 is the peak load of the cleaning equipment, in kilowatt (kW); Ra3 is the storage capacity of the cleaning fluid, in liter (L); Ra4 is the equipment operation stability, ranging from 0 to 1; Ra5 is the utilization ratio of renewable resources in the cleaning process, ranging from 0 to 1; Ra6 is the equipment response time, in seconds (s); Ra7 is the temperature of the equipment operating environment, in degrees Celsius (°C); Ra8 is the maintenance cost of the equipment, in RMB (yuan); Ra9 is the carbon emissions generated during the cleaning process, in grams (g); Ra10 is user satisfaction, ranging from 0 to 10.

[0030] 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 data attributes may reach dozens or even hundreds.

[0031] In this embodiment, the collected data can be labeled. The labeling method of the present invention is manual labeling. In one embodiment, the labeled categories can include three categories: "high efficiency", "standard" and "low efficiency".

[0032] Step S104, performing data expansion on the data based on a generative adversarial network of quantum entangled states to obtain expanded data.

[0033] In the task of the present invention, the acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor generalization ability of the model and affect the accuracy of the model. The present invention uses a generative adversarial network based on quantum entanglement to generate samples, thereby achieving data expansion.

[0034] In some embodiments, the above-mentioned generative adversarial network of quantum entangled state includes a generator and a discriminator, and the input layer of the generator adopts the representation of quantum state.

[0035] The quantum entangled state-based generative adversarial network in this embodiment includes a generator and a discriminator. On the basis of the traditional generative adversarial network, the input layer of the generator adopts the representation of quantum states, and improves the intrinsic correlation and expression ability of the data by simulating the characteristics of quantum entanglement, so as to solve the problem of insufficient sample diversity in the traditional generative adversarial network.

[0036] In some embodiments, the generative adversarial network trains the generator and the discriminator through the following steps: the generative adversarial network is initialized; in each iteration, the generator receives the quantum state representation of random noise and target data, simulates the generation of entangled states through quantum gate operations, and outputs data samples; the discriminator obtains real samples and data samples output by the generator, and updates the weights of the discriminator based on the posterior probabilities of the real samples and the data samples output by the generator; back propagates based on the gradient calculated based on a preset loss function, and updates the parameters and quantum state parameters of the generative adversarial network; determines whether the preset stop iteration condition is met; if the stop iteration condition is met, stops the iteration and determines the final generative adversarial network; if the stop iteration condition is not met, continues to iterate.

[0037] Specifically, the training process of the quantum entangled state-based generative adversarial network algorithm is shown in steps A1-A5: Step A1, let the generator be , the discriminator is , set the initialization parameters and quantum state parameters of the generative adversarial network based on quantum entanglement In one embodiment, the initialization method is expressed as:

[0038] In the formula, It means that it obeys a specific distribution. represents a normal distribution, represents the initial quantum state parameters, is the initialization variance, represents the initial network weights.

[0039] Step A2: In each iteration, the generator accepts the quantum state representation of random noise and target data, simulates the generation of entangled state through quantum gate operation, and outputs new data samples. The conversion operation of quantum state is expressed as:

[0040] In the formula, is the quantum state conversion function, is the random noise of the input, is an imaginary unit, is the number of qubits, It is caused by noise The parameterized Quantum state angles, Indicates The ground state of a quantum bit.

[0041] Furthermore, the quantum state angle reflects the influence of input noise on the quantum state angle, and the calculation method is expressed as:

[0042] In the formula, , , , , is a learnable parameter used to adjust the noisy data The effect on the quantum state angle, thus optimizing the diversity and quality of data generation, specifically, is the learnable parameter of the first quantum state angle, is the learnable parameter of the second quantum state angle, is the learnable parameter from the third quantum state angle, is the learnable parameter of the fourth quantum state angle, is the learnable parameter of the fifth quantum state angle; is the hyperbolic tangent function.

[0043] Furthermore, the generator outputs corresponding data samples according to the quantum state, which will be used in the training process of the discriminator. The generation process is expressed as:

[0044] In the formula, represents the generated data sample, is a generator function, is the input quantum state, are the weights of the generative adversarial network.

[0045] In step A3, the discriminator receives the generated data samples and the real samples, calculates the posterior probabilities of the two, updates their weights, and optimizes the ability to judge authenticity, which is expressed as:

[0046] In the formula, is the weight The posterior probability of is a real data sample, is the likelihood probability of the data under given weights, is the prior probability of the weight, is the prior probability of the data.

[0047] In one embodiment, the calculation method of the likelihood probability of data under a given weight is expressed as:

[0048] In the formula, represents an exponential function with a natural constant as base, represents the L2 norm.

[0049] Step A4, back propagation is performed according to the gradient calculated by the loss function to update the parameters of the generative adversarial network and the quantum state parameters to optimize the performance of the entire network. The parameter update method is expressed as:

[0050] In the formula, Indicates a parameter update operation. is the learning rate of the generative adversarial network, is the loss function of the generative adversarial network, is the loss function over weights The gradient of is the loss function for the quantum state parameters Preferably, Set to 0.01.

[0051] In one embodiment, the loss function of the generative adversarial network is calculated as:

[0052] In the formula, is the coefficient that balances the impact of positive and negative samples, is the regularization coefficient. Preferably, Set to 0.3.

[0053] Furthermore, taking the weight update process as an example, the calculation method of the gradient of the loss function to the weight is expressed as:

[0054] In the formula, It represents the gradient of the discriminator output with respect to the weights, which characterizes how the discriminator distinguishes between generated data and real data based on the current weights.

[0055] Step A5, 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.

[0056] After the data expansion model training is completed, the trained data expansion model is used to increase the number of samples. In one embodiment, assuming that the original collected samples are 800, and the data expansion model expands and generates 200 samples, the expanded data set contains 1000 samples.

[0057] Step S106, training a feature extraction model based on simulated load balancing based on the expanded data, and the feature extraction model outputs the data after feature extraction; wherein, the weight update in each training cycle of the feature extraction model is adaptively adjusted through a dynamic variable weight strategy.

[0058] The expanded data is input into the feature extraction model for training the feature extraction model. The present invention uses a 6-layer fully connected neural network for feature extraction. In the prior art, some solutions use neural networks for feature extraction. In some neural network structures, gradient vanishing, gradient explosion or falling into local optimal solutions may be encountered, affecting the stability of training and the performance of the model. The present invention uses a neural network algorithm based on simulated load balancing optimization as a feature extraction model, and through a dynamic variable weight strategy, the weight update in each training cycle can be adaptively adjusted continuously to improve the feature adaptive refinement capability of feature extraction.

[0059] In some embodiments, the weights and biases of the network layer of the feature extraction model can be initialized; at the beginning of each training batch, the weights of the network layer of the feature extraction model are dynamically adjusted based on simulated load balancing according to the training results of the previous batch; when the data stream passes through the network of the feature extraction model for feature extraction, based on a preset load balancing optimization algorithm, the training speed and accuracy of the feature extraction model are dynamically adjusted using the response time and energy consumption of the current network; the network structure of the feature extraction model is adjusted based on the features extracted by the feature extraction model, and the network parameters of the feature extraction model are corrected based on the contribution weights; it is determined whether the preset stop iteration condition is met; if the stop iteration condition is met, the iteration is stopped and the final feature extraction model is determined; if the stop iteration condition is not met, the iteration continues.

[0060] Specifically, the training process of the neural network algorithm based on simulated load balancing optimization is shown in steps B1-B5: Step B1, initializing the weights and biases of all neural network layers. In one embodiment, for the neural network layer Layer weights and bias , the initialization method is expressed as:

[0061] In the formula, For the neural network The weight of the layer, For the neural network The bias of the layer, is the number of neurons in the previous layer, represents a normal distribution with mean 0 and variance 1.

[0062] Step B2, at the beginning of each training batch, the weights of the network layer are dynamically adjusted according to the training results of the previous batch. The process of neural network parameters simulates the load balancing optimization problem. Considering the dynamics of load changes, the calculation method of the parameter update amount is expressed as:

[0063] In the formula, For the neural network The weight update amount of the layer, is the learning rate of the neural network, is the loss function of the neural network, is the sparsity regularization coefficient, is a sign function. Preferably, Set to 0.3.

[0064] In one embodiment, taking weight parameter update as an example, the calculation method of the weight gradient of the loss function of the neural network is expressed as:

[0065] In the formula, is the output of the l-1th layer of the neural network, The neural network The error of the layer, for The transpose of .

[0066] Furthermore, the neural network The error of a layer is calculated by back propagating the weights and errors of the next layer, expressed as:

[0067] here, is the activation function in The derivative of the layer, The neural network The net input of the layer, The neural network The weight of the layer, is element-wise multiplication, The neural network Layer error.

[0068] Step B3, when the data stream passes through the network for feature extraction, according to the load balancing optimization algorithm, the training speed and accuracy are dynamically adjusted using the current network response time and energy consumption to ensure that energy efficiency is maximized without sacrificing model performance. Specifically, the load balancing optimization algorithm is implemented by adjusting the batch size and learning rate of the training, and the calculation method is expressed as:

[0069] In the formula, The number of samples to be input to the network for the next batch, To obtain the maximum value function, Indicates the current The training accuracy of the iteration represents the processing power of the model in the balanced optimization algorithm; Indicates the current The number of batch samples in the iteration, which represents the system load of the model in the balanced optimization algorithm; is the initial learning rate of the neural network, is the learning rate of the neural network in the th iteration, Indicates the current iteration number. is the learning rate decay coefficient, is the load regulation coefficient. Preferably, Set to 1, Set to 0.05.

[0070] In one embodiment, the learning rate decay coefficient depends on the rate of change of the system load and is calculated as:

[0071] In the formula, To find the differential function, is the learning rate decay adjustment coefficient. Preferably, Set to 3.

[0072] Step B4, use the extracted features to fine-tune the network structure to better adapt to data changes. Considering the importance and contribution of data features, the weight correction based on contribution is used to update the neural network parameters. Taking the weight parameters as an example, the update method is expressed as:

[0073] In the formula, is the learning rate adjustment coefficient of the neural network weight parameter, is the strength parameter of the sparse regularization, The neural network is Tier The weight of a neuron, is a stable decimal value to avoid division by zero. Preferably, Set to 0.1, Set to 0.0001.

[0074] In one embodiment, the strength parameter of the sparse regularization is updated by adaptive adjustment. The adaptive mechanism allows the model to dynamically adjust the sparse regularization strength according to the current feature sparsity, expressed as:

[0075] In the formula, is the sparse regularization adjustment rate, is the target sparsity level, It is The sparsity objective function of the layer. Preferably, Set to 2, Set to 0.5.

[0076] Further, The calculation method of the sparsity objective function of the layer is expressed as:

[0077] In the formula, It is The sparsity objective function of the layer is used to measure the sparsity of the layer output; The neural network Tier The input features of neurons, The neural network is Tier The weight of a neuron.

[0078] Step B5, 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.

[0079] Step S108, training a feature dimensionality reduction model based on the data after feature extraction, and the feature dimensionality reduction model outputs the reduced-dimensional data; wherein the feature dimensionality reduction model is determined based on an autoencoder algorithm of a Gaussian process latent variable.

[0080] The data after feature extraction is input into the feature dimensionality reduction model for training the feature dimensionality reduction model. The present invention adopts an autoencoder algorithm based on Gaussian process latent variables as the feature dimensionality reduction model. The Gaussian process is listened to in the hidden layer of the autoencoder. The structure of the latent space is optimized by simulating the distribution of input data, thereby improving the generalization ability of the model and enhancing the model's ability to capture the intrinsic structure of the data.

[0081] In some embodiments, the parameters of the autoencoder and the Gaussian process can be initialized; the encoder transforms the data after feature extraction to obtain a representation of the latent space; the Gaussian process models the distribution of the latent variables; sparse matrix decomposition is used in the latent space to reduce the data dimension; back propagation is performed according to the loss function of the autoencoder to adjust the parameters of the feature dimensionality reduction model; it is determined whether a preset stop iteration condition is met; if the stop iteration condition is met, the iteration is stopped and the final feature dimensionality reduction model is determined; if the stop iteration condition is not met, the iteration continues.

[0082] Specifically, the training process of the autoencoder algorithm based on Gaussian process latent variables is shown in steps C1-C5: Step C1, initialize the relevant parameters of the autoencoder and Gaussian process, and the initialization method is expressed as:

[0083] In the formula, For the autoencoder The weight of the layer; For the autoencoder Bias of the layer; is the number of neurons in the previous layer; is the kernel function parameter of the Gaussian process; is uniformly distributed in the interval [0,1].

[0084] In step C2, the input feature-extracted data is transformed through the encoder part to obtain the representation of the latent space. The Gaussian process is used to model the distribution of the latent variables to ensure the continuity and smoothness of the latent space. The calculation method of the encoding and Gaussian process parts is expressed as:

[0085] In the formula, is the latent variable after coding; is the ReLU activation function; and are the weights and biases of the encoding layer respectively; The feature vector after input feature extraction, that is, the input vector of the autoencoder; and They are the mean function and covariance function in the Gaussian process, based on the latent variables calculate; is a Gaussian process function.

[0086] In one embodiment, the covariance function in the Gaussian process determines the smoothness and flexibility of the Gaussian process, and the calculation method is expressed as:

[0087] In the formula, is the variance of the kernel function, which determines the amplitude of the function value change; is the length parameter of the kernel function, which determines the smoothness of the function; is the L2 norm; To distinguish Another potential variable.

[0088] Step C3, sparse matrix decomposition is used in the latent space to further reduce the data dimension by retaining the most important features and ignoring redundancy and noise. The calculation method is expressed as:

[0089] In the formula, Indicates that the input matrix is ​​decomposed into singular values, retaining the previous Large singular values ​​enable sparse representation; The number of singular values ​​to be retained is determined based on data complexity and processing requirements.

[0090] Furthermore, in the process of singular value decomposition, the method of selecting singular values ​​is expressed as:

[0091] In the formula, is the left singular vector obtained by singular value decomposition, is the singular value diagonal matrix obtained by singular value decomposition, is the right singular vector obtained by singular value decomposition, is the transpose of the right singular vector obtained by singular value decomposition; The number of singular values ​​selected determines the amount of information retained in the data after dimensionality reduction.

[0092] Step C4, back propagation is performed according to the loss function of the autoencoder, and the model parameters are adjusted to minimize the total loss. The calculation method of the weight update amount of the autoencoder is expressed as:

[0093] In the formula, is the loss function of the autoencoder. In one embodiment, the reconstruction error loss is used as the loss function of the autoencoder. is the regularization loss of the Gaussian process; is the learning rate of the autoencoder; is the regularization coefficient of the autoencoder, controlling the impact of Gaussian process loss. Preferably, Set to 0.01.

[0094] Furthermore, the calculation method of the gradient of the loss function to the weight is expressed as:

[0095] In the formula, is the decoded output, for The transpose of is the encoded output; Represents the trace operation, which is used to calculate the contribution of the kernel matrix gradient to the weight gradient. is the kernel matrix, its gradient Depends on the specific parameters of the Gaussian process.

[0096] Step C5, 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.

[0097] Step S110, training a classifier model based on the data after dimensionality reduction; wherein a deep neural decision tree algorithm based on adaptive weight sharing is adopted as the classification algorithm of the classifier model.

[0098] The reduced-dimensional data is input into the classifier to train the classifier model. The present invention adopts a deep neural decision tree algorithm based on adaptive weight sharing as the classification algorithm, and utilizes the nonlinear modeling ability of deep learning and the interpretability of the decision tree to achieve efficient classification and decision-making. The present invention implements weight sharing on each node of the neural decision tree, dynamically adjusts the sharing level according to the characteristics of the input data, enhances the adaptability of the model to different data distributions, and improves the efficiency and stability of model training. In some embodiments, the structure and node parameters of the deep neural decision tree can be initialized; wherein the parameters of each deep neural decision tree node include decision weights and biases; output the output value of each node through the forward propagation process; dynamically adjust the weight sharing strategy according to the flow of samples in the decision tree; evaluate the complexity and performance of the classifier model during the training process, and dynamically adjust the depth and number of nodes of the decision tree; determine whether the preset stop iteration condition is met; if the stop iteration condition is met, stop the iteration and determine the final classifier model; if the stop iteration condition is not met, continue to iterate.

[0099] Specifically, the training process of the deep neural decision tree algorithm based on adaptive weight sharing is shown in steps D1-D5: Step D1, initializing the structure and node parameters of the deep neural decision tree. The parameters of each node include decision weight and bias. In one embodiment, the initialization method is expressed as:

[0100] In the formula, Represents the initialized deep neural decision tree weights; Represents the initialized deep neural decision tree bias; and Represents the number of nodes in the previous layer and the current layer respectively; Represents the data after feature dimensionality reduction, that is, the feature matrix input to the deep neural decision tree.

[0101] Step D2, the forward propagation process includes calculating the output value of each decision node. Among them, each input sample calculates the decision result of each input sample at each node according to the current weight and bias, and determines the path of each input sample in the decision tree. The forward propagation process is expressed as:

[0102] In the formula, For the deep neural decision tree Linear transformation result of the layer; is the result calculated by the activation function; and For the deep neural decision tree Layer weights and biases; For the deep neural decision tree The input characteristics of the layer; is the Sigmoid activation function.

[0103] Step D3: dynamically adjust the weight sharing strategy according to the flow of samples in the decision tree. For features that frequently appear in the same path, enhance their weight sharing.

[0104] Furthermore, the Jacobi iterative solution method is used to calculate the gradient of the loss function with respect to each node parameter, and then back-propagation is performed to update the parameters. The calculation method of the updated amount of weights and biases of the deep neural decision tree is expressed as:

[0105] In the formula, and They are the first The amount of update to the layer’s weights and biases; is the learning rate of the deep neural decision tree; is the loss function of the deep neural decision tree; and They are the gradients of the loss function of the deep neural decision tree with respect to weights and biases, respectively.

[0106] Furthermore, the loss function of the deep neural decision tree is calculated using information entropy loss, which is expressed as:

[0107] In the formula, represents the normalized class probability, is the output category.

[0108] Step D4, evaluates the complexity and performance of the model during the training process, and dynamically adjusts the depth and number of nodes of the decision tree as needed to prevent overfitting and optimize performance. For overly complex model structures, a dynamic pruning mechanism is used to optimize model performance and prevent overfitting. The calculation method is expressed as:

[0109] In the formula, The function indicates whether the absolute value of the weight is greater than the threshold To decide whether to keep the weight; represents element-wise multiplication; The importance criterion for deciding the weights is the pruning threshold.

[0110] Step D5, 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.

[0111] Step S112, evaluating the energy efficiency of the cleaning equipment based on the trained feature extraction model, feature dimension reduction model and classifier model.

[0112] The model of the present invention is applied to the energy efficiency evaluation and optimization of cleaning equipment, and the trained model is used to predict and classify new input data, thereby realizing accurate evaluation and optimization of energy efficiency.

[0113] In some embodiments, input data of a cleaning device can be obtained; the input data is input into a trained feature extraction model to output the input data after feature extraction; the input data after feature extraction is input into a trained feature dimension reduction model to output the input data after dimension reduction; the input data after dimension reduction is input into a trained classifier model to output the energy efficiency of the cleaning device.

[0114] The original input data collected in this embodiment is first input into the trained feature extraction and feature dimension reduction model for feature processing. Further, the processed features are input into the classifier model for classifier training to finally obtain the classification result.

[0115] Among them, the classification results are divided into the following three categories: "efficient", "standard" and "inefficient".

[0116] Determine operational optimization suggestions for the cleaning equipment based on the energy efficiency of the cleaning equipment; wherein the operational optimization suggestions include at least one of the following: adjusting operating parameters of the cleaning equipment, optimizing the ratio of cleaning agent use, or improving equipment operation logic.

[0117] Furthermore, based on the classification results, the system provides targeted operational optimization suggestions, such as: adjusting the operating parameters of the cleaning equipment, such as spray pressure, flow rate and temperature settings; optimizing the cleaning agent usage ratio to improve cleaning efficiency; improving equipment operation logic to reduce unnecessary energy consumption.

[0118] In addition, by predicting the maintenance needs of equipment, potential faults can be discovered in advance, the occurrence of sudden problems can be reduced, maintenance costs can be reduced, and the reliability and economy of equipment operation can be improved.

[0119] In summary, an embodiment of the present invention provides an artificial intelligence-based cleaning equipment energy efficiency evaluation and optimization method, including: obtaining data based on log records during the operation of the cleaning equipment, and labeling the data; wherein the labeled categories are used to characterize the energy efficiency of the cleaning equipment; performing data expansion on the data based on a generative adversarial network of quantum entangled states to obtain expanded data; training a feature extraction model based on simulated load balancing based on the expanded data, and the feature extraction model outputs the feature extracted data; wherein the weight update in each training cycle of the feature extraction model is adaptively adjusted through a dynamic variable weight strategy; training a feature dimension reduction model based on the feature extracted data, and the feature dimension reduction model outputs the reduced dimension data; wherein the feature dimension reduction model is determined based on an autoencoder algorithm of a Gaussian process latent variable; training a classifier model based on the reduced dimension data; wherein a deep neural decision tree algorithm based on adaptive weight sharing is used as the classification algorithm of the classifier model; and evaluating the energy efficiency of the cleaning equipment based on the trained feature extraction model, feature dimension reduction model and classifier model.

[0120] The above method adopted by the embodiment of the present invention has the following advantages: 1. Through data expansion based on quantum entangled state generative adversarial network, the characteristics of quantum entanglement are simulated to improve the intrinsic correlation and expression ability of data, and solve the problem of insufficient sample diversity in traditional generative adversarial network.

[0121] 2. A fully connected neural network optimized with simulated load balancing is used as the feature extraction model. The dynamic variable weight strategy is used to adaptively adjust the weight update in each training cycle, solving the problems of gradient vanishing, gradient exploding, or falling into the local optimal solution that traditional neural networks may encounter in feature extraction.

[0122] 3. The autoencoder based on Gaussian process latent variables is used to achieve feature dimensionality reduction. The structure of the latent space is optimized by simulating the distribution of input data, which solves the problem that traditional feature dimensionality reduction technology may ignore the intrinsic structure and distribution of data.

[0123] 4. A deep neural decision tree based on adaptive weight sharing is used as the classifier, and the weight sharing level is dynamically adjusted according to the characteristics of the input data, which solves the problems of low efficiency and poor model generalization ability of traditional classifiers when processing different data distributions.

[0124] Corresponding to the above method embodiment, this embodiment provides an artificial intelligence-based cleaning equipment energy efficiency evaluation and optimization device, which is implemented on the basis of the above embodiment, see Figure 2 The schematic diagram of the structure of a cleaning equipment energy efficiency evaluation and optimization device based on artificial intelligence is shown, and the cleaning equipment energy efficiency evaluation and optimization device based on artificial intelligence includes: The data acquisition and labeling module 21 is used to acquire data based on the log records during the operation of the cleaning equipment and label the data; wherein the labeled category is used to characterize the energy efficiency of the cleaning equipment; A data expansion module 22, used for performing data expansion on the data based on a generative adversarial network of quantum entangled states to obtain expanded data; The feature extraction model training module 23 is used to train the feature extraction model based on the simulated load balancing based on the expanded data, and the feature extraction model outputs the data after feature extraction; wherein the weight update in each training cycle of the feature extraction model is adaptively adjusted through a dynamic variable weight strategy; A feature dimensionality reduction model training module 24 is used to train a feature dimensionality reduction model based on the data after feature extraction, and the feature dimensionality reduction model outputs the data after dimensionality reduction; wherein the feature dimensionality reduction model is determined based on an autoencoder algorithm of a Gaussian process latent variable; The classifier model training module 25 is used to train the classifier model based on the data after dimensionality reduction; wherein a deep neural decision tree algorithm based on adaptive weight sharing is used as the classification algorithm of the classifier model; The cleaning equipment energy efficiency evaluation module 26 is used to evaluate the energy efficiency of the cleaning equipment based on the trained feature extraction model, feature dimension reduction model and classifier model.

[0125] The embodiment of the present invention provides an energy efficiency evaluation and optimization device for cleaning equipment based on artificial intelligence. Data expansion is performed through a generative adversarial network based on quantum entangled states, and the characteristics of quantum entanglement are simulated to enhance the intrinsic correlation and expression ability of data, thereby solving the problem of insufficient sample diversity in traditional generative adversarial networks. A fully connected neural network with simulated load balancing optimization is adopted as a feature extraction model, and a dynamic variable weight strategy is utilized to adaptively adjust the weight update in each training cycle, thereby solving the problems of gradient vanishing, gradient exploding, or falling into a local optimal solution that may be encountered by traditional neural networks in feature extraction. An autoencoder based on Gaussian process latent variables is adopted to realize feature dimensionality reduction, and the structure of the latent space is optimized by simulating the distribution of input data, thereby solving the problem that traditional feature dimensionality reduction technology may ignore the intrinsic structure and distribution of data. A deep neural decision tree based on adaptive weight sharing is adopted as a classifier, and the weight sharing level is dynamically adjusted according to the characteristics of the input data, thereby solving the problems of low efficiency and poor model generalization ability of traditional classifiers when processing different data distributions.

[0126] The above-mentioned generative adversarial network of quantum entangled state includes a generator and a discriminator, and the input layer of the generator adopts the representation of quantum state.

[0127] See also Figure 3 The schematic diagram of the structure of another cleaning equipment energy efficiency evaluation and optimization device based on artificial intelligence is shown, and the above-mentioned cleaning equipment energy efficiency evaluation and optimization device based on artificial intelligence also includes: a generator and a discriminator training module 27, which is used to initialize the generative adversarial network; in each iteration, the generator receives the quantum state representation of random noise and target data, simulates the generation of entangled states through quantum gate operations, and outputs data samples; the discriminator obtains real samples and data samples output by the generator, and updates the weights of the discriminator based on the posterior probabilities of the real samples and the data samples output by the generator; back propagates based on the gradient calculated based on the preset loss function, and updates the parameters and quantum state parameters of the generative adversarial network; determines whether the preset stop iteration condition is met; if the stop iteration condition is met, stops the iteration and determines the final generative adversarial network; if the stop iteration condition is not met, continues to iterate.

[0128] The feature extraction model training module is used to initialize the weights and biases of the network layer of the feature extraction model; at the beginning of each training batch, the weights of the network layer of the feature extraction model are dynamically adjusted based on the simulated load balancing according to the training results of the previous batch; when the data stream passes through the network of the feature extraction model for feature extraction, based on the preset load balancing optimization algorithm, the training speed and accuracy of the feature extraction model are dynamically adjusted using the response time and energy consumption of the current network; the network structure of the feature extraction model is adjusted based on the features extracted by the feature extraction model, and the network parameters of the feature extraction model are corrected based on the weight of the contribution; it is determined whether the preset stop iteration condition is met; if the stop iteration condition is met, the iteration is stopped and the final feature extraction model is determined; if the stop iteration condition is not met, the iteration is continued.

[0129] The feature dimension reduction model training module is used to initialize the parameters of the autoencoder and the Gaussian process; the encoder transforms the data after feature extraction to obtain the representation of the latent space; the Gaussian process models the distribution of the latent variables; sparse matrix decomposition is used in the latent space to reduce the data dimension; back propagation is performed according to the loss function of the autoencoder to adjust the parameters of the feature dimension reduction model; it is determined whether the preset stop iteration condition is met; if the stop iteration condition is met, the iteration is stopped and the final feature dimension reduction model is determined; if the stop iteration condition is not met, the iteration is continued.

[0130] The above-mentioned classifier model training module is used to initialize the structure and node parameters of the deep neural decision tree; wherein the parameters of each deep neural decision tree node include decision weight and bias; output the output value of each node through the forward propagation process; dynamically adjust the weight sharing strategy according to the flow of samples in the decision tree; evaluate the complexity and performance of the classifier model during the training process, and dynamically adjust the depth and number of nodes of the decision tree; determine whether the preset stop iteration condition is met; if the stop iteration condition is met, stop the iteration and determine the final classifier model; if the stop iteration condition is not met, continue to iterate.

[0131] The above-mentioned classifier model training module is used to calculate the decision result of each input sample at each node according to the current weight and bias for each input sample, and determine the path of each input sample in the decision tree.

[0132] The above-mentioned cleaning equipment energy efficiency evaluation module is used to obtain the input data of the cleaning equipment; input the input data into the trained feature extraction model, and output the input data after feature extraction; input the input data after feature extraction into the trained feature dimension reduction model, and output the input data after dimension reduction; input the input data after dimension reduction into the trained classifier model, and output the energy efficiency of the cleaning equipment.

[0133] like Figure 3As shown, the above-mentioned artificial intelligence-based cleaning equipment energy efficiency evaluation and optimization device also includes: an operation optimization suggestion module 28, which is used to determine the operation optimization suggestion of the cleaning equipment based on the energy efficiency of the cleaning equipment; wherein the operation optimization suggestion includes at least one of the following: adjusting the operating parameters of the cleaning equipment, optimizing the cleaning agent usage ratio or improving the equipment operation logic.

[0134] Technical personnel in the relevant field can clearly understand that, for the convenience and conciseness of description, the specific working process of the above-described artificial intelligence-based cleaning equipment energy efficiency evaluation and optimization system can refer to the corresponding process in the aforementioned embodiment of the artificial intelligence-based cleaning equipment energy efficiency evaluation and optimization method, and will not be repeated here.

[0135] The embodiment of the present invention also provides an electronic device for running the above-mentioned cleaning equipment energy efficiency evaluation and optimization method based on artificial intelligence; see Figure 4 A structural schematic diagram of an electronic device is shown, which includes a memory 100 and a processor 101, wherein the memory 100 is used to store one or more computer instructions, and the one or more computer instructions are executed by the processor 101 to implement the above-mentioned artificial intelligence-based cleaning equipment energy efficiency evaluation and optimization method.

[0136] Further, Figure 4 The electronic device shown further includes a bus 102 and a communication interface 103 , and the processor 101 , the communication interface 103 and the memory 100 are connected via the bus 102 .

[0137] The memory 100 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk storage. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 103 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used. The bus 102 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0138] The processor 101 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 101. The above processor 101 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in conjunction with the embodiments of the present invention can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module may be located in a storage medium mature in the art, such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 100, and the processor 101 reads the information in the memory 100 and completes the steps of the method of the above embodiment in combination with its hardware.

[0139] An embodiment of the present invention also provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are called and executed by a processor, the computer-executable instructions prompt the processor to implement the above-mentioned artificial intelligence-based cleaning equipment energy efficiency evaluation and optimization method. The specific implementation can be found in the method embodiment, which will not be repeated here.

[0140] The artificial intelligence-based cleaning equipment energy efficiency evaluation and optimization method and device, and computer program product provided in the embodiments of the present invention include a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods in the previous method embodiments. For specific implementations, please refer to the method embodiments, which will not be repeated here.

[0141] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system and / or device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0142] In addition, in the description of the embodiments of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0143] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0144] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0145] Finally, it should be noted that the above embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for energy efficiency evaluation and optimization of cleaning equipment based on artificial intelligence, characterized in that: The method comprises: Acquire data based on log records during the operation of the cleaning device, and annotate the data; wherein the annotated category is used to characterize the energy efficiency of the cleaning device; Performing data expansion on the data based on a generative adversarial network of quantum entangled states to obtain expanded data; Based on the expanded data, a feature extraction model based on simulated load balancing is trained, and the feature extraction model outputs the data after feature extraction; wherein the weight update in each training cycle of the feature extraction model is adaptively adjusted through a dynamic variable weight strategy; A feature dimension reduction model is trained based on the data after feature extraction, and the feature dimension reduction model outputs the reduced-dimensional data; wherein the feature dimension reduction model is determined based on an autoencoder algorithm of a Gaussian process latent variable; A classifier model is trained based on the reduced-dimensional data; wherein a deep neural decision tree algorithm based on adaptive weight sharing is used as the classification algorithm of the classifier model; The energy efficiency of the cleaning equipment is evaluated based on the trained feature extraction model, the feature dimension reduction model and the classifier model.

2. The method according to claim 1, characterized in that: The generative adversarial network of the quantum entangled state includes a generator and a discriminator, and the input layer of the generator adopts the representation form of the quantum state.

3. The method according to claim 2, characterized in that The generative adversarial network trains the generator and the discriminator through the following steps: Initializing the generative adversarial network; In each iteration, the generator receives the quantum state representation of random noise and target data, simulates the generation of entangled state through quantum gate operation, and outputs data samples; The discriminator obtains a real sample and a data sample output by the generator, and updates a weight of the discriminator based on the posterior probability of the real sample and the data sample output by the generator; Back propagation is performed based on the gradient calculated by the preset loss function to update the parameters of the generative adversarial network and the quantum state parameters; Determine whether a preset stop iteration condition is met; if the stop iteration condition is met, stop iteration and determine the final generative adversarial network; if the stop iteration condition is not met, continue iterating.

4. The method according to claim 1, characterized in that: The step of training a feature extraction model based on simulated load balancing based on the expanded data comprises: Initialize the weights and biases of the network layers of the feature extraction model; At the beginning of each training batch, dynamically adjusting the weights of the network layers of the feature extraction model based on simulated load balancing according to the training results of the previous batch; When the data stream passes through the network of the feature extraction model for feature extraction, based on a preset load balancing optimization algorithm, the training speed and accuracy of the feature extraction model are dynamically adjusted using the response time and energy consumption of the current network; Adjusting the network structure of the feature extraction model based on the features extracted by the feature extraction model, and correcting the network parameters of the feature extraction model based on the weight of the contribution; Determine whether a preset stop iteration condition is met; if the stop iteration condition is met, stop iteration and determine the final feature extraction model; if the stop iteration condition is not met, continue iteration.

5. The method according to claim 1, characterized in that The step of training a feature dimension reduction model based on the feature-extracted data includes: Initialize the parameters of the autoencoder and Gaussian process; The encoder transforms the feature-extracted data to obtain a representation of the latent space; the Gaussian process models the distribution of the latent variables; Using sparse matrix factorization in the latent space to reduce the dimensionality of the data; Performing back propagation according to the loss function of the autoencoder to adjust the parameters of the feature dimensionality reduction model; Determine whether a preset stop iteration condition is met; if the stop iteration condition is met, stop iteration and determine the final feature dimension reduction model; if the stop iteration condition is not met, continue iteration.

6. The method according to claim 1, characterized in that The step of training a classifier model based on the dimensionally reduced data comprises: Initialize the structure and node parameters of the deep neural decision tree; wherein the parameters of each node of the deep neural decision tree include decision weights and biases; Output the output value of each node through the forward propagation process; A strategy for dynamically adjusting weight sharing according to the flow of samples in the decision tree; Evaluating the complexity and performance of the classifier model during training, and dynamically adjusting the depth and number of nodes of the decision tree; Determine whether a preset stop iteration condition is met; if the stop iteration condition is met, stop iteration and determine the final classifier model; if the stop iteration condition is not met, continue iteration.

7. The method according to claim 6, characterized in that The steps of outputting the output value of each node through the forward propagation process include: For each input sample, a decision result of each input sample at each node is calculated according to the current weight and bias, and a path of each input sample in the decision tree is determined.

8. The method according to claim 1, characterized in that: The step of evaluating the energy efficiency of the cleaning equipment based on the trained feature extraction model, the feature dimension reduction model and the classifier model comprises: Obtaining input data of the cleaning device; Inputting the input data into the trained feature extraction model, and outputting the input data after feature extraction; Inputting the input data after feature extraction into the trained feature dimension reduction model, and outputting the input data after dimension reduction; The input data after dimension reduction is input into the trained classifier model, and the energy efficiency of the cleaning equipment is output.

9. The method according to any one of claims 1 to 8, characterized in that: After the step of evaluating the energy efficiency of the cleaning equipment based on the trained feature extraction model, the feature dimension reduction model and the classifier model, the method further includes: Determine an operation optimization suggestion for the cleaning equipment based on the energy efficiency of the cleaning equipment; wherein the operation optimization suggestion includes at least one of the following: adjusting the operation parameters of the cleaning equipment, optimizing the ratio of cleaning agent use, or improving the equipment operation logic.

10. An artificial intelligence-based cleaning equipment energy efficiency evaluation and optimization device, characterized in that: The device comprises: A data acquisition and labeling module, used to acquire data based on log records during the operation of the cleaning equipment and label the data; wherein the labeled category is used to characterize the energy efficiency of the cleaning equipment; A data expansion module, used for performing data expansion on the data based on a generative adversarial network of quantum entangled states to obtain expanded data; A feature extraction model training module, used for training a feature extraction model based on simulated load balancing based on the expanded data, the feature extraction model outputting the data after feature extraction; wherein the weight update in each training cycle of the feature extraction model is adaptively adjusted through a dynamic variable weight strategy; A feature dimensionality reduction model training module, used for training a feature dimensionality reduction model based on the data after feature extraction, wherein the feature dimensionality reduction model outputs the reduced-dimensional data; wherein the feature dimensionality reduction model is determined based on an autoencoder algorithm of a Gaussian process latent variable; A classifier model training module, used for training a classifier model based on the reduced-dimensional data; wherein a deep neural decision tree algorithm based on adaptive weight sharing is used as the classification algorithm of the classifier model; The cleaning equipment energy efficiency evaluation module is used to evaluate the energy efficiency of the cleaning equipment based on the trained feature extraction model, the feature dimension reduction model and the classifier model.

Images