Multi-user heterogeneous load identification method and system

Abstract

The invention discloses a multi-user heterogeneous load identification method and system. The method comprises the following steps: acquiring first data of target electric equipment, and establishing an initial load sample library according to the first data; pre-training a plurality of first models and a plurality of second models, and performing first cooperative training on all the first models and the second models to obtain a third model; and performing real-time load identification according to the third model in combination with the initial load sample library. The cooperative training method does not need to limit the completely consistent structure of each user model, and can fully cooperate with the multi-user heterogeneous model to improve the trained effect; the effect of knowledge distillation for cooperative training of the multi-user load identification model is effectively improved; the designed improved convolution module can realize high-accuracy load identification under the condition of extremely few parameters and calculation amount, the lightweight degree of a compressed model in knowledge distillation is effectively improved, and the practicability of cooperative training is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of multi - user heterogeneous load identification, and particularly to a multi - user heterogeneous load identification method and system. Background Art

[0002] Load identification is an important technical means to analyze the internal electricity consumption behavior and equipment status of users from bus data using intelligent algorithms. Through load identification technology, grid operators can monitor and analyze the electricity consumption of users without deploying a large number of device - level sensors, thereby improving the transparency and efficiency of grid operation. With the rapid development of the power system and the popularization of electric energy substitution applications, the accuracy and model adaptability of load identification have received increasing attention. The proposed multi - user collaborative load identification model provides a new solution to solve the problem of insufficient single - user data and improve the generalization ability of the model. However, the existing multi - user collaborative training methods still face many challenges. Generally speaking, the existing load identification models still have the following main problems:

[0003] (1) The existing collaborative training methods for multi - user load identification models force the load identification model structures of different users to be exactly the same, with a high degree of constraint, and cannot adapt to the development trend of personalized load identification models for users;

[0004] (2) The existing load identification method based on knowledge distillation only aligns the compressed model with the original model through the load identification probability vector output by the model, and cannot be used to simultaneously utilize global generality information and local specificity information during collaborative training, nor is it easy to effectively align in the heterogeneity of load identification models and the differences in user electricity consumption data, resulting in poor collaborative training effects;

[0005] (3) The existing load identification models have limited feature extraction capabilities and are difficult to achieve high - accuracy load identification with extremely few parameters and computational amounts, resulting in limited lightweight of the compressed models selected in knowledge distillation and affecting the efficiency of collaborative training of multi - user load identification models. Summary of the Invention

[0006] The purpose of this part is to outline some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this part, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this part, the abstract, and the title. However, such simplifications or omissions cannot be used to limit the scope of the present invention.

[0007] In view of the above - mentioned existing problems, the present invention is proposed.

[0008] Therefore, the present invention provides a multi - user heterogeneous load identification method and system, which can solve the problems mentioned in the background art.

[0009] To solve the above technical problems, the present invention provides the following technical solutions:

[0010] In a first aspect, the present invention provides a multi-user heterogeneous load recognition method, including:

[0011] Obtain the first data of the target electrical equipment, and establish an initial load sample library according to the first data;

[0012] Pre-train a number of first models and a number of second models, perform first collaborative training on all the first models and the second models, and obtain a third model;

[0013] The first model and the third model are used for heterogeneous load recognition, and the second model is used for collaborative training of the first model;

[0014] Perform real-time load recognition according to the third model in combination with the initial load sample library.

[0015] As a preferred solution of the multi-user heterogeneous load recognition method of the present invention, wherein: the first model and the third model include:

[0016] The first model is any model whose input is the first data in the initial load sample library and whose output is the type of electrical equipment in the initial load sample library or relevant parameters that can directly or indirectly obtain the type of electrical equipment;

[0017] The third model is the first model after the first collaborative training.

[0018] As a preferred solution of the multi-user heterogeneous load recognition method of the present invention, wherein: the first collaborative training includes:

[0019] Use the second model to perform collaborative training on the first model;

[0020] The input and output of the second model are the same as the input and output of the first model, and there are several identical structures between the first model and the second model.

[0021] As a preferred solution of the multi-user heterogeneous load recognition method of the present invention, wherein: the first collaborative training further includes:

[0022] Align the first model and the second model at the same time;

[0023] Perform an alignment operation on the first model towards the second model;

[0024] The alignment operation refers to minimizing the relative entropy of the load recognition probability vectors between the first model and the second model, and maximizing the mutual information of the outputs of the penultimate fully connected layers of the first model and the second model.

[0025] As a preferred solution of the multi-user heterogeneous load identification method described in the present invention, wherein: the last two fully connected layers of the first model and the second model are the same.

[0026] As a preferred solution of the multi-user heterogeneous load identification method described in the present invention, wherein: the obtaining of the first data of the target electrical equipment and the establishment of the initial load sample library according to the first data include:

[0027] The first data of the target electrical equipment includes at least several groups of voltage and current signals, and the phases of the voltage and current signals are kept aligned;

[0028] The initial load sample library includes at least several groups of voltage and current sequences of power frequency and the types of electrical equipment.

[0029] As a preferred solution of the multi-user heterogeneous load identification method described in the present invention, wherein: the second model includes several deformable convolution modules and fully connected layers;

[0030] A single deformable convolution module consists of cascaded deformable convolution calculations and activation function calculations.

[0031] In a second aspect, the present invention provides a multi-user heterogeneous load identification system, including:

[0032] A sample library establishment module, configured to obtain the first data of the target electrical equipment and establish an initial load sample library according to the first data;

[0033] A collaborative training module, configured to pre-train several first models and several second models, perform first collaborative training on all the first models and the second models, and obtain a third model;

[0034] The first model and the third model are used for heterogeneous load identification, and the second model is used for collaborative training of the first model;

[0035] A load identification module, configured to perform real-time load identification according to the third model in combination with the initial load sample library.

[0036] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the method described above are implemented.

[0037] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described above are implemented.

[0038] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention proposes a multi-user heterogeneous load recognition method and system, which acquires first data of a target electrical device and establishes an initial load sample library according to the first data; pre-trains a number of first models and a number of second models, and performs first collaborative training on all the first models and the second models to obtain a third model; the first model and the third model are used for heterogeneous load recognition, and the second model is used for collaborative training of the first model; real-time load recognition is performed according to the third model in combination with the initial load sample library. The collaborative training method that does not require the structures of each user model to be exactly the same can fully collaborate multiple user heterogeneous models to improve the training effect; effectively improve the effect of knowledge distillation for collaborative training of multi-user load recognition models; the designed improved convolution module can achieve high-accuracy load recognition with extremely few parameters and computational amounts, effectively improving the lightweight degree of the compressed model in knowledge distillation and improving the practicality of collaborative training. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts. Among them:

[0040] Figure 1 is a method flow chart of a multi-user heterogeneous load recognition method and system provided by an embodiment of the present invention;

[0041] Figure 2 is an internal structure diagram of a computer device of a multi-user heterogeneous load recognition method and system provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following will give a detailed description of the specific embodiments of the present invention with reference to the drawings of the specification. Obviously, the described embodiments are some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0043] Embodiment 1

[0044] Refer to Figure 1 - Figure 2 , which is the first embodiment of the present invention. This embodiment provides a multi-user heterogeneous load recognition method and system, including:

[0045] In the existing related technologies, there are some problems. For example, the collaborative training method of the load recognition model forcibly requires that the structures of the load recognition models of different users are exactly the same, which results in a high degree of model constraint and is difficult to adapt to the development trend of the personalization of the user load recognition model. In addition, the existing load recognition method based on knowledge distillation only aligns the compressed model with the original model through the load recognition probability vector output by the model. This method cannot effectively utilize both the global generality information and the local specificity information in collaborative training, and it is also difficult to effectively align in the heterogeneity of the load recognition model and the differences in user electricity consumption data, thus affecting the effect of collaborative training. Moreover, the existing load recognition model has limited feature extraction ability and is difficult to achieve high-accuracy load recognition with limited parameters and computational complexity. This limits the lightweight degree of the compressed model selected in knowledge distillation and further affects the efficiency of the collaborative training of the multi-user load recognition model.

[0046] This application provides a method that can effectively solve the above-mentioned problems. Next, multiple embodiments will be combined to elaborate in detail how to implement the multi-user heterogeneous load recognition method;

[0047] Figure 1 A method flowchart of a multi-user heterogeneous load recognition method and system is shown, including:

[0048] S101, obtain the first data of the target electrical equipment, and establish an initial load sample library according to the first data;

[0049] In an optional embodiment, the target electrical equipment may include various household appliances, industrial equipment, or other power consumption equipment.

[0050] In an optional embodiment, the manner of obtaining the target electrical equipment may include, but is not limited to, directly obtaining through sensors, smart meters, or other data acquisition devices, or obtaining from a remote data source through a communication network.

[0051] In an optional embodiment, the manner of establishing the initial load sample library may include collecting a large amount of voltage and current signal data from the target electrical equipment. These data need to ensure that the phases of the voltage and current signals are aligned to ensure the accuracy and consistency of the data. Then, preprocess these data, extract several groups of voltage and current sequences of the power frequency cycle, and combine with the type information of the electrical equipment to construct the initial load sample library. This sample library will serve as the basis for subsequent model training and load recognition.

[0052] It should be noted that the initial load sample library established in this way can fully reflect the load characteristics of different electrical equipment and provide strong data support for subsequent multi-user heterogeneous load recognition.

[0053] In the embodiments of the present application, obtaining the first data of the target electrical equipment and establishing an initial load sample library according to the first data includes:

[0054] The first data of the target electrical equipment includes at least several groups of voltage and current signals, and the phases of the voltage and current signals are kept aligned;

[0055] The initial load sample library includes at least several groups of voltage and current sequences of a cycle and the type of electrical equipment.

[0056] In an alternative embodiment, each user respectively collects high-frequency voltage and current signals of the electrical equipment based on intelligent sensors to establish an initial load sample library. The frequencies of the high-frequency voltage and current signals are not less than 1 kHz, and the phases of the voltage and current signals are kept aligned;

[0057] In the embodiments of the present application, a single sample in the initial load sample library is composed of voltage and current sequences of not less than 100 cycles and the corresponding electrical equipment category information, and the number of samples in the load sample library of each user is not less than 50.

[0058] In an alternative embodiment, the form of the initial load sample library may include but is not limited to databases, data tables, data files, etc., for facilitating subsequent data processing and model training. The data in the initial load sample library should ensure accuracy and integrity to ensure the effect of model training and the accuracy of load identification. In addition, the initial load sample library needs to be updated and maintained regularly to adapt to changes in different electrical equipment and load characteristics.

[0059] It should be noted that in this way, a comprehensive, accurate and reliable initial load sample library can be established, providing strong data support for subsequent multi-user heterogeneous load identification.

[0060] It should also be noted that obtaining the first data of the target electrical equipment and establishing an initial load sample library according to the first data can ensure the accuracy and reliability of subsequent model training and load identification. As the basis for model training and load identification, the quality and integrity of the initial load sample library directly affect the final effect. By establishing a comprehensive, accurate and reliable initial load sample library, strong data support can be provided for subsequent multi-user heterogeneous load identification, thereby improving the accuracy and efficiency of load identification. In addition, a high-quality initial load sample library can also help the model better adapt to the load characteristics and electricity consumption behaviors of different users, further enhancing the generalization ability and practicality of the model.

[0061] S102. Pretrain several first models and several second models, and perform first collaborative training on all the first models and second models to obtain a third model;

[0062] In the embodiments of the present application, the first model and the third model are used for heterogeneous load identification, and the second model is used for co-training the first model;

[0063] In the embodiments of the present application, the first model and the third model include:

[0064] The first model is any model with the first data in the initial load sample library as the input and the electrical equipment type in the initial load sample library or the relevant parameters that can directly or indirectly obtain the electrical equipment type as the output;

[0065] The third model is the first model after the first co-training.

[0066] In the embodiments of the present application, the last two fully connected layers of the first model and the second model are the same.

[0067] In an alternative embodiment, the first model can be various neural network models based on deep learning algorithms, such as convolutional neural network (CNN), recurrent neural network (RNN), or long short-term memory network (LSTM), etc. These models have powerful feature extraction and pattern recognition capabilities, and can automatically learn and extract effective load features from the input voltage and current signals, so as to accurately identify the electrical equipment type.

[0068] In an alternative embodiment, the second model can also be a neural network model based on deep learning algorithms, but its structure and parameters can be different from those of the first model. The main role of the second model is to be the object of co-training, and by interacting and co-training with the first model, it helps the first model improve the accuracy and generalization ability of load identification.

[0069] In an alternative embodiment, the first model can also be other types of machine learning models, such as support vector machine (SVM), decision tree, random forest, etc. These models may have unique advantages and effects in different application scenarios, and through co-training, they can complement the first model and jointly improve the performance of load identification.

[0070] In the embodiments of the present application, each user constructs a user heterogeneous load identification model (i.e., the first model) and a homogeneous compressed model (i.e., the second model), and performs supervised training based on the above initial load sample library. The homogeneous compressed models refer to that the structures of the compressed models in each user are the same, and each is composed of num dc deformable convolution modules and num fc fully connected layers stacked together. In a specific implementation, num dc and num fc can be selected as 3 and 2 respectively; the calculation method of the fully connected layer is shown in the following formula:

[0071] Among them

[0072] In the formula is the r-th element in the output vector, is the s-th element in the input vector, W rs is the element in the r-th row and s-th column of the weight matrix, b r is the r-th element in the bias vector, and are the number of rows and columns of the weight vector respectively. In a specific implementation and can both be set to 128;

[0073] In an alternative embodiment, a single deformable convolution module consists of cascaded deformable convolution calculations and activation function calculations. The deformable convolution calculation is as shown in the following formula:

[0074]

[0075] In the formula is the calculation result at the p0-th position of the deformable convolution, is the number of positions involved in a single deformable convolution calculation, Δp k is the k-th position offset, is the position offset of the k-th position participating in training, is the weight value at the k-th position in the deformable convolution calculation, is the input at the -th position in the input of the deformable convolution calculation, and ensures that the position offset is an integer by rounding down;

[0076] In an alternative embodiment, the activation function calculation is as shown in the following formula:

[0077]

[0078] In the formula y act is the output of the activation function calculation, x act is the input of the activation function calculation, and a is the slope weight participating in model training in the activation function calculation;

[0079] In an alternative embodiment, a heterogeneous load recognition model refers to a load recognition model whose structure is different from that of the homogeneous compressed model except for the last two fully connected layers being the same;

[0080] In an alternative embodiment, the inputs of the heterogeneous load recognition model and the homogeneous compressed model are both voltage-current trajectories, where the voltage sequence is and the current sequence is Plot the corresponding voltage and current as coordinate points in a plane rectangular coordinate system to form a trajectory representing the relationship between voltage and current;

[0081] In an optional embodiment, supervised training refers to updating the weights based on the obtained initial load sample library to minimize the cross-entropy loss function. The cross-entropy loss function Loss ce is shown as follows:

[0082]

[0083] where o is the total number of samples in a single initial load sample library, VI l is the voltage-current trajectory of the l-th sample, t real (·) and q pred (·) respectively represent the actual load recognition probability vector and the load recognition probability vector output by the load recognition model. The actual load recognition probability vector is a vector with a dimension value of 1 corresponding to the electrical equipment category and a value of 0 for the remaining dimensions.

[0084] It should be noted that through the above method, several first models and several second models can be pre-trained, and through the first collaborative training of these models, a third model with stronger load recognition ability can be obtained. During the first collaborative training process, the first model and the second model will learn from each other and optimize each other, so that the third model shows higher accuracy and generalization ability in load recognition.

[0085] In the embodiment of the present application, the first collaborative training includes:

[0086] Using the second model to perform collaborative training on the first model;

[0087] The input and output of the second model are the same as those of the first model, and there are several identical structures between the first model and the second model.

[0088] In the embodiment of the present application, the first collaborative training further includes:

[0089] Aligning the first model and the second model simultaneously;

[0090] Performing an alignment operation on the first model towards the second model;

[0091] The alignment operation refers to minimizing the relative entropy of the load recognition probability vectors between the first model and the second model, and maximizing the mutual information of the outputs of the penultimate fully connected layers of the first model and the second model.

[0092] In an optional embodiment, the first co-training can be achieved through bidirectional knowledge distillation co-training, or by means of joint optimization. Bidirectional knowledge distillation co-training means that the first model and the second model learn each other's knowledge during the training process, that is, the output of one model is used as the soft label of the other model for training, so as to improve their respective load recognition capabilities. Joint optimization means jointly optimizing the parameters of the first model and the second model, so that the load recognition performance of both is jointly improved during the co-training process.

[0093] In an optional embodiment, in order to further improve the accuracy and efficiency of load recognition, post-processing can also be performed on the third model. The post-processing methods can include but are not limited to smoothing the output of the third model, filtering, or making comprehensive judgments by combining other feature information. Through post-processing, noise and interference can be further eliminated, and the stability and reliability of load recognition can be improved.

[0094] It should be noted that the multi-user heterogeneous load recognition system constructed in the above manner can accurately identify various types of electrical equipment and adapt to the load characteristics and electricity consumption behaviors of different users. This system not only has the advantages of high precision and high efficiency, but also can be flexibly configured and expanded according to the needs of different users, and has good practicability and application value.

[0095] In the embodiment of the present application, each user is managed by a cloud server, and the user heterogeneous load recognition models of each user are co-trained through bidirectional knowledge distillation with the compressed isomorphic models.

[0096] In the embodiment of the present application, the second model includes several deformable convolution modules and a fully connected layer;

[0097] A single deformable convolution module consists of cascaded deformable convolution calculations and activation function calculations.

[0098] In the embodiment of the present application, each user is managed by a cloud server, and the user heterogeneous load recognition models of each user are co-trained through bidirectional knowledge distillation with the compressed isomorphic models. The cloud server manages each user and is composed of a multi-user access module and a multi-user co-training module. The multi-user access module has functions of network security authentication for accessing users, recording user information, and monitoring the network operation status of users; the multi-user co-training module is responsible for collecting the compressed isomorphic models of each user, calculating the average value of the weights of the compressed isomorphic models, and then distributing them to each user for bidirectional knowledge distillation. The above co-training process is iterated num fl times. In a specific implementation, num fl can be set to 100;

[0099] In an alternative embodiment, bidirectional knowledge distillation refers to simultaneously aligning the homogeneous compressed model with the heterogeneous load recognition model and aligning the heterogeneous load recognition model with the homogeneous compressed model. Alignment means minimizing the relative entropy of the load recognition probability vectors between the homogeneous compressed model and the heterogeneous load recognition model, and maximizing the mutual information of the output of the penultimate fully connected layer of the homogeneous compressed model to the heterogeneous load recognition model. The loss function Loss for minimizing the relative entropy of the load recognition probability vectors between the homogeneous compressed model and the heterogeneous load recognition model is as follows kl As shown in the following formula:

[0100]

[0101] where o is the total number of samples in a single initial load sample library, and VI l is the voltage-current trajectory of the l-th sample, toral (·) and qcom (·) represent the load recognition probability vector of the heterogeneous load recognition model and the load recognition probability vector of the homogeneous compressed model respectively;

[0102] In an alternative embodiment, according to the Donsker-Varadhan theorem, the loss function Loss for maximizing the mutual information of the output of the penultimate fully connected layer of the homogeneous compressed model to the heterogeneous load recognition model is as follows dv As shown in the following formula:

[0103]

[0104] where z oral and z com are the output vectors of the penultimate fully connected layers of the heterogeneous load recognition model and the homogeneous compressed model respectively, and F is a function with an upper bound in the mapping from the output vector of the penultimate fully connected layer of the model to the real number domain.

[0105] It should be noted that other theorems such as Jensen's Inequality or Mutual Information Neural Estimation (MINE) can also be used to define and calculate the loss function for maximizing mutual information. These theorems and methods have extensive applications in the fields of machine learning and information theory and can effectively measure and optimize the information transmission and sharing between models.

[0106] It should also be noted that pre-training a number of first models and a number of second models, and performing first co-training on all the first models and second models to obtain a third model can significantly improve the load recognition ability of the third model. First of all, the pre-training process can provide a good initialization state for the model, enabling it to quickly adapt to subsequent training tasks. Through pre-training, the model can learn some basic features and rules, laying a solid foundation for subsequent co-training. Secondly, the first co-training process enables the first model and the second model to learn from each other and optimize each other. Through methods such as knowledge distillation and joint optimization, the load recognition performance of the models is jointly improved during the co-training process. This mutual learning mechanism helps the model capture more features and details, thereby improving the accuracy and generalization ability of load recognition. Finally, by post-processing the third model, noise and interference can be further eliminated, improving the stability and reliability of load recognition. Therefore, the pre-training and co-training processes are of great significance for constructing a high-precision and high-efficiency multi-user heterogeneous load recognition system.

[0107] S103, perform real-time load recognition according to the third model in combination with the initial load sample library.

[0108] In an optional embodiment, the real-time load recognition process may include inputting the voltage-current trajectory to be recognized into the trained third model, obtaining a load recognition probability vector through the forward propagation calculation of the model, and determining the final load category according to the probability vector. To further improve the efficiency and accuracy of real-time load recognition, optimization techniques such as parallel computing, model pruning, and quantization can be adopted to reduce the computational amount and storage requirements of the model. At the same time, the model can also be optimized and adjusted specifically in combination with the actual electricity consumption scenario and load characteristics to meet the needs and electricity consumption behaviors of different users. Through real-time load recognition, the system can accurately monitor and classify various types of electrical equipment, providing strong support for subsequent energy management and optimization.

[0109] Exemplarily, during the real-time load recognition process, the system can compare the recognition result with the threshold or rule set by the user, thereby triggering corresponding energy management strategies. For example, when the system detects that the load of a certain electrical equipment is abnormal or exceeds the preset power limit, it can automatically send an alarm or adjust the power supply strategy to prevent energy waste or equipment failure. In addition, the system can also provide personalized energy usage suggestions and optimization plans for users based on the real-time load recognition results, helping them better manage and control energy consumption and achieve the goal of energy conservation and emission reduction.

[0110] When the first threshold and the second threshold are set, if the probability value of a certain electrical equipment category in the load recognition probability vector of the voltage-current trajectory to be recognized exceeds the first threshold, it is determined that the electrical equipment is in the on state;

[0111] If the probability value of a certain type of electrical equipment is lower than the second threshold, it is determined that the electrical equipment is in the off state;

[0112] If the probability value of a certain type of electrical equipment is between the first threshold and the second threshold, it is determined that the electrical equipment is in the standby state or the load fluctuation state.

[0113] It should be noted that through such threshold setting and determination rules, the system can more precisely monitor and control the state of electrical equipment, improving the accuracy and efficiency of energy management.

[0114] At the same time, users can also flexibly adjust the settings of the first threshold and the second threshold according to actual needs to adapt to different electrical usage scenarios and load characteristics.

[0115] In summary, the present invention proposes a multi-user heterogeneous load recognition method, which obtains the first data of the target electrical equipment and establishes an initial load sample library according to the first data; pre-trains a number of first models and a number of second models, and performs first collaborative training on all the first models and the second models to obtain a third model; the first model and the third model are used for heterogeneous load recognition, and the second model is used for collaborative training of the first model; real-time load recognition is performed according to the third model in combination with the initial load sample library. A collaborative training method that does not require the model structures of each user to be exactly the same can fully collaborate multiple user heterogeneous models to improve the post-training effect; effectively improve the effect of knowledge distillation for collaborative training of multi-user load recognition models; the designed improved convolution module can achieve high-accuracy load recognition with extremely few parameters and computational amounts, effectively improving the lightweight degree of the compressed model in knowledge distillation and improving the practicality of collaborative training.

[0116] Embodiment 2

[0117] In a preferred embodiment, to more specifically illustrate the implementation manner of the present invention, this embodiment provides an actual application scenario of a multi-user heterogeneous load recognition system. In this scenario, the system is deployed in a power system including multiple users, and each user is equipped with a smart meter or other power monitoring devices for real-time collection of voltage and current data. These data are used as inputs and processed and analyzed by the multi-user heterogeneous load recognition system provided by the present invention.

[0118] In specific implementation, the system first establishes an initial load sample library according to the collected voltage and current data. Then, these sample libraries are used to pre-train multiple first models and second models. The purpose of pre-training is to provide a good initialization state for the models so that they can quickly adapt to subsequent training tasks. Through pre-training, the models can learn some basic features and laws, such as the change trends of voltage and current, the periodic changes of load, etc.

[0119] After the pre-training is completed, the system starts the first collaborative training on the first model and the second model. In this stage, the first model and the second model will learn from and optimize each other. Specifically, the system will use the output of one model as the soft label of the other model for training, so as to improve their respective load recognition capabilities. At the same time, the system will also calculate the average value of the model weights after isomorphic compression and distribute it to each user for two-way knowledge distillation. This process will be iterated multiple times until the load recognition performance of the model reaches a stable state.

[0120] After the collaborative training, the system obtains a third model with stronger load recognition ability. This model can accurately identify various types of electrical equipment and adapt to the load characteristics and electricity consumption behaviors of different users. In practical applications, the system will perform real-time load recognition according to the third model combined with the initial load sample library. Specifically, the system will collect the voltage and current data of users in real time, input these data into the third model, and then output the load recognition result.

[0121] To further improve the accuracy and efficiency of load recognition, the system can also post-process the output of the third model. The post-processing methods can include but are not limited to smoothing the output, filtering the output, or making comprehensive judgments by combining other feature information, etc. Through post-processing, noise and interference can be further eliminated, and the stability and reliability of load recognition can be improved.

[0122] In summary, this embodiment provides a practical application scenario of a multi-user heterogeneous load recognition system. This system can accurately identify various types of electrical equipment and adapt to the load characteristics and electricity consumption behaviors of different users. At the same time, this system also has the advantages of high precision, high efficiency, flexible configuration and expansion, etc., and has good practicality and application value.

[0123] Embodiment 3

[0124] This embodiment also provides a multi-user heterogeneous load recognition system, including:

[0125] A sample library establishment module, configured to obtain the first data of the target electrical equipment and establish an initial load sample library according to the first data;

[0126] A collaborative training module, configured to pre-train a number of first models and a number of second models, perform the first collaborative training on all the first models and the second models, and obtain a third model;

[0127] The first model and the third model are used for heterogeneous load recognition, and the second model is used for collaborative training of the first model;

[0128] A load identification module, configured to perform real-time load identification according to a third model in combination with an initial load sample library.

[0129] Each of the above unit modules may be embedded in a processor in a computer device in hardware form or independent of the processor, or may be stored in a memory in the computer device in software form, so that the processor can call and execute the operations corresponding to each of the above modules.

[0130] This embodiment also provides a computer device, which may be a terminal, and its internal structure diagram may be as Figure 2 shown. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a multi-user heterogeneous load identification method. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the outer shell of the computer device, or an external keyboard, a touchpad, or a mouse, etc.

[0131] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0132] Obtain first data of a target power-consuming device, and establish an initial load sample library according to the first data;

[0133] Pre-train a number of first models and a number of second models, perform first collaborative training on all the first models and the second models, and obtain a third model;

[0134] The first model and the third model are used for heterogeneous load identification, and the second model is used for collaborative training of the first model;

[0135] Perform real-time load identification according to the third model in combination with the initial load sample library.

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

[0137] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming languages such as Java and interpreted scripting languages such as JavaScript.

[0138] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, as well as the combination of flows and / or blocks in the flowchart and / or block diagram. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0139] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1Steps of the functions specified in one or more boxes.

[0141] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications to these embodiments once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications falling within the scope of the present application.

[0142] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these modifications and variations.

Claims

1. A multi-user heterogeneous load identification method, characterized in that: include: Acquire first data of the target electrical equipment, and establish an initial load sample library according to the first data; Pre-training a plurality of first models and a plurality of second models, performing first collaborative training on all the first models and the second models, and obtaining a third model; The first model and the third model are used for heterogeneous load identification, and the second model is used for collaborative training of the first model; Real-time load identification is performed according to the third model in combination with the initial load sample library.

2. The multi-user heterogeneous load identification method according to claim 1, characterized in that: The first model and the third model include: The first model is any model whose input is the first data in the initial load sample library and whose output is the type of electric equipment in the initial load sample library or the relevant parameters of the type of electric equipment that can be directly or indirectly obtained; The third model is the first model after the first collaborative training.

3. The multi-user heterogeneous load identification method according to claim 2, characterized in that: The first collaborative training includes: Co-training the first model using the second model; The input and output of the second model are the same as the input and output of the first model, and the first model and the second model have some common structures.

4. The multi-user heterogeneous load identification method according to claim 3, characterized in that: The first collaborative training also includes: Aligning the first model and the second model simultaneously; Aligning the first model to the second model; The alignment operation refers to minimizing the relative entropy of the load recognition probability vector between the first model and the second model, and maximizing the mutual information between the outputs of the first model and the penultimate fully connected layer of the second model.

5. The multi-user heterogeneous load identification method according to claim 4, characterized in that: The last two fully connected layers of the first model and the second model are the same.

6. The multi-user heterogeneous load identification method according to claim 5, characterized in that: The step of acquiring first data of the target electrical equipment and establishing an initial load sample library according to the first data comprises: The first data of the target electrical device includes at least a plurality of groups of voltage and current signals, and the phases of the voltage and current signals are aligned; The initial load sample library includes at least several groups of cycle voltage and current sequences and types of electrical equipment.

7. The multi-user heterogeneous load identification method according to claim 6, characterized in that: The second model includes a plurality of variable convolution modules and a fully connected layer; A single deformable convolution module consists of cascaded deformable convolution calculations and activation function calculations.

8. A multi-user heterogeneous load identification system, characterized in that: include: A sample library establishment module, used to obtain first data of a target electrical device and establish an initial load sample library according to the first data; A collaborative training module, used for pre-training a plurality of first models and a plurality of second models, performing a first collaborative training on all the first models and the second models, and obtaining a third model; The first model and the third model are used for heterogeneous load identification, and the second model is used for collaborative training of the first model; The load identification module is used to perform real-time load identification based on the third model in combination with the initial load sample library.

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

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

Images