House energy consumption non-intrusive load monitoring method and device and storage medium

Abstract

The invention discloses a non-intrusive load monitoring method and device for house energy consumption and a storage medium, and belongs to the technical field of power monitoring, and the method comprises the following steps: carrying out the random initialization of all particles through employing a dynamic particle swarm optimization algorithm, and obtaining a particle hyper-parameter combination; generating a long short-term memory network model; dividing the obtained historical power classification data set into a training set and a test set according to timestamps, training through the training set, and calculating through the test set to obtain a root-mean-square difference; using the root-mean-square difference as an adaptive value to update the individual optimal solution of the particles and the current global optimal solution of the whole particle swarm, then calculating the weight of the particles, and updating the speed and the position according to the weight of the particles; determining the optimal solution of the iteration complete bureau as the optimal parameter configuration of the long short-term memory network model; and inputting the real-time power data to obtain a load monitoring result. The method can solve the problems that an existing non-intrusive load monitoring method is high in complexity, long in consumed time and low in monitoring result accuracy.

Description

Technical Field

[0001] The invention relates to a non-intrusive load monitoring method, device and storage medium for house energy consumption, and belongs to the technical field of power monitoring. Background Art

[0002] Energy demand has grown significantly over the past 40 years. More and more electronic devices have become indispensable in homes; computers, tablets, mobile phones, etc. This, of course, means that more energy is consumed. Therefore, it is necessary to effectively monitor the energy consumed. With the advancement of technology, it is already possible to use smart meters to monitor the energy consumption of a house. This approach is to install a smart meter in each appliance in the home and obtain real-time information about the energy consumption. This way of using smart meters at the level of household appliances is still quite expensive, so other ways must be explored.

[0003] The existing technology adopts a cost-effective method to monitor the energy consumption of houses. Specifically, each household only needs to install one meter to monitor the total energy consumed. This is possible in the real world with the help of algorithms that can infer the total energy signal into the device-level sub-signals that constitute the signal. This is also called the energy decomposition task, and the field of research on this task is non-intrusive load monitoring (NILM). The information obtained by non-intrusive load monitoring has great practical value for all electricity users. For ordinary users, if the device details obtained through load monitoring can be fed back in time, it will help guide users to use electricity reasonably, thereby achieving energy conservation and consumption reduction, and saving electricity bills. For power companies, non-intrusive load monitoring can achieve fine-grained perception of each component of the load without significantly increasing investment, improve the accuracy of load forecasting of the power system, improve the safety and economy of the power grid, and help to more accurately model user behavior and realize differentiated and precise services for users. For power equipment manufacturers, the equipment status and corresponding energy consumption information provided by non-intrusive load monitoring can provide a basis for equipment fault diagnosis or further realize predictive maintenance of equipment. Non-intrusive load monitoring is an interdisciplinary application of computer, communication, electronics and electrical engineering, and can be widely used in building energy conservation, smart cities, smart grids and other fields.

[0004] The most popular non-intrusive load monitoring (NILM) method in machine learning is to generate models. The training input is high-frequency or low-frequency data of the house power supply, and the label is the actual energy consumption value of the selected device. Usually, the training data and labels are arranged in chronological order, and one of the goals of the selected model is to find the correlation between the aggregate signal and the device signal within a given time range. In the prior art, non-intrusive load monitoring uses a particle swarm optimization algorithm (PSO) combined with a long short-term memory network (LSTM), in which these particles continuously collaborate to find the best LSTM configuration through the update iteration of the particle swarm speed and the information sharing of the particle swarm. However, there are the following problems: First, the restriction range and even type of hyperparameters are different, and the particles in the traditional PSO cannot undertake this task. Unreasonable constraints may cause the model to be overfitted or underfitted, thereby increasing the convergence time, extending the monitoring calculation time and reducing the accuracy of the monitoring results. In addition, the traditional PSO is often easy to stop searching when it finds the local optimum, so it misses the global optimal position, which will affect the accuracy of the final monitoring results. Finally, when using PSO to adjust the LSTM parameters, the calculation of the particle fitness value requires model training, resulting in a long monitoring calculation process. Summary of the invention

[0005] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide a non-invasive load monitoring method, device and storage medium for house energy consumption, which can solve the problems of high complexity, long time consumption and low accuracy of monitoring results in the existing non-invasive load monitoring methods.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0007] In a first aspect, the present invention provides a non-intrusive load monitoring method for house energy consumption, comprising the following steps:

[0008] Use the dynamic particle swarm optimization algorithm to randomly initialize all particles and obtain the particle hyperparameter combination;

[0009] Start the iterative process to generate a LSTM network model based on the particle hyperparameter combination;

[0010] The acquired historical power classification data set is divided into a training set and a test set according to the timestamp, the long short-term memory network model is trained with the training set, and the root mean square error is calculated with the test set;

[0011] The root mean square error is used as the fitness value to update the individual optimal solution of the particle and the current global optimal solution of the entire particle swarm. Then the weight of the particle is calculated, and the speed and position are updated according to the weight of the particle to enter a new round of iteration.

[0012] Repeat the above iterative process until the real-time iteration number reaches the preset maximum iteration number, determine the current global optimal solution at this time as the optimal parameter configuration of the long short-term memory network model, and obtain the long short-term memory network model with the optimal parameters;

[0013] The real-time power data is input into the parameter-optimal long short-term memory network model to obtain the load monitoring results.

[0014] The dynamic particle swarm optimization algorithm uses mixed variable encoding. Each particle represents a hyperparameter combination of the long short-term memory network. Each hyperparameter is designed with a corresponding variable. Each particle has four adjustable hyperparameter variables and their corresponding speeds.

[0015] The long short-term memory network model includes two hidden layers, a fully connected layer and a softmax layer. The first hidden layer receives input data, extracts timing features from the input data through its gating units and cell states, and outputs them to the next hidden layer; the second hidden layer receives the output from the first hidden layer, and continues to extract more abstract timing features, and then passes the timing features to the fully connected layer. The fully connected layer outputs data to the softmax layer. The softmax layer is used for classification tasks and maps the output of the fully connected layer to a probability distribution.

[0016] The historical power classification dataset includes ground device plug loads for various devices, and two fully labeled device event timestamps.

[0017] The historical electricity classification dataset was cleaned, deduplicated, normalized, feature engineered, and transformed before use.

[0018] The labels in the historical power classification dataset and real-time power data are five digits, where the first three digits indicate the type of equipment and the last two digits indicate the operation mode.

[0019] The weight of a particle is calculated as follows: At each iteration, the algorithm evaluates the fitness of each particle i and obtains the fitness of all particles , where n is the total number of particles, and then compare all , determine the maximum fitness value and the minimum fitness value respectively and , calculate the weight of the i-th particle The formula is:

[0020] (1)

[0021] (2)

[0022] (3)

[0023] in, and Represent the maximum weight value and the minimum weight value respectively. and Represents the maximum and current value of the number of iterations, is an intermediate variable used for calculation, Represents a random number that obeys a uniform distribution, ranging from [0,1]. Each particle i will have an independent random number.

[0024] A master process and n slave processes are created through the distributed computing framework, where n is the number of particles. The particle hyperparameter combination is sent from the master process to the slave process through the remote function of the distributed computing framework, and the root mean square error is returned from the slave process to the master process.

[0025] In a second aspect, the present invention provides a non-intrusive load monitoring device for house energy consumption, comprising:

[0026] The particle initialization module is used to randomly initialize all particles using the dynamic particle swarm optimization algorithm to obtain a combination of particle hyperparameters;

[0027] The model generation module is used to start the iterative process and generate a long short-term memory network model based on the particle hyperparameter combination;

[0028] A root mean square error acquisition module is used to divide the acquired historical power classification data set into a training set and a test set according to timestamps, train the long short-term memory network model through the training set, and then calculate the root mean square error through the test set;

[0029] The particle update module uses the root mean square error as the fitness value to update the individual optimal solution of the particle and the current global optimal solution of the entire particle swarm. It then calculates the particle weight, updates the speed and position according to the particle weight, and enters a new round of iteration.

[0030] The model optimal parameter determination module is used to repeat the above iterative process until the real-time iteration number reaches the preset maximum iteration number, determine the current global optimal solution at this time as the optimal parameter configuration of the long short-term memory network model, and obtain the long short-term memory network model with the optimal parameters;

[0031] The load monitoring result acquisition module is used to input real-time power data into the parameter-optimal long short-term memory network model to obtain load monitoring results.

[0032] In a third aspect, the present invention provides a computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the steps of the non-intrusive load monitoring method for house energy consumption.

[0033] Beneficial effects of the present invention: The present invention provides a non-invasive load monitoring method and system for house energy consumption, which is based on a dynamic particle swarm optimization algorithm and a long short-term memory network, and uses the root mean square error as an adaptation value to update the individual optimal solution of the particle and the current global optimal solution of the entire particle swarm, and then calculates the weight of the particle, and updates the speed and position according to the weight of the particle, and changes the weight of the dynamic particle swarm optimization algorithm, which helps to jump out of the possible local optimal position and determine the global optimal position. In addition, it can solve the constraint problem of hyperparameters, make the model fitting degree reasonable, and reduce the convergence time. In summary, it can improve the final detection prediction result, and can improve the search efficiency and reduce the time consumption.

[0034] By adopting a distributed framework, one master process and n slave processes are created through the distributed computing framework. The particle hyperparameter combination is sent from the master process to the slave process through the remote function of the distributed computing framework, and the root mean square error is returned from the slave process to the master process. This can reduce the model training time when calculating the particle fitness value and reduce the time-consuming monitoring calculation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 Schematic diagram of the RNN structure;

[0036] Figure 2 Schematic diagram of the LSTM structure of the long short-term memory neural network;

[0037] Figure 3 A flow chart of a non-intrusive load monitoring method for house energy consumption according to the present invention;

[0038] Figure 4 The accuracy corresponding to different iteration times when the refrigerator of the present invention is in the open state and the closed state (EMBED data set);

[0039] Figure 5 The accuracy corresponding to different iteration times of the kettle in the open state and the closed state of the present invention (EMBED data set);

[0040] Figure 6 It is the accuracy rate corresponding to different iteration numbers when the bread maker of the present invention is in the open state and the closed state (EMBED data set). DETAILED DESCRIPTION

[0041] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the protection scope of the present invention.

[0042] Example 1

[0043] The present invention discloses a non-intrusive load monitoring method for house energy consumption, based on (dynamic particle swarm optimization algorithm) DPSO and (long short-term memory network) LSTM hyperparameter tuning, the method optimizes the selection of hyperparameters such as the optimal number of hidden units, activation function and learning rate in LSTM, and finds a high-performance LSTM model for non-intrusive monitoring, such as Figure 3 As shown, the following steps are included:

[0044] Step 1, particle initialization: All particles are randomly initialized by the DPSO algorithm. The hyperparameter combination of each particle is randomly assigned, and n+1 processes are created, namely n slave processes and 1 master process, where n is the number of particles. DPSO is an optimization method for continuous nonlinear functions. DPSO constructs particles to simulate individuals in a population, and all particles constitute a population, that is, a particle swarm. Particles share their individual optimal solutions with the particle swarm, and then the best individual optimal solution among all individuals can be found through information sharing among all particles in the entire particle swarm. This solution is the current global optimal solution in the entire particle swarm. Through continuous iteration, the entire particle swarm will gradually approach the global optimal solution of the entire search space through dynamic changes.

[0045] In the present invention, DPSO adopts mixed variable encoding to solve the hyperparameter constraint problem, and designs the corresponding speed and position update process. Each particle represents a hyperparameter combination of LSTM, so it is necessary to design a corresponding variable for each hyperparameter. Each particle has 4 adjustable hyperparameter variables and their corresponding speeds, and their different meanings are given in Table 1.

[0046] Table 1 Hybrid particle encoding scheme

[0047]

[0048] Step 2: Use the remote function of the distributed framework (Ray) to send the particle hyperparameter combination in the current iteration to the slave process for building and training the LSTM model. LSTM is a variant of RNN (recurrent neural network). Figure 1 It can be seen that the characteristic of RNN is that its structure has chain-like repeated modules. The structure of the repeated module of LSTM has changed compared with RNN, which is also its improvement over RNN. Unlike RNN, which only has one tanh layer inside the module, LSTM has a four-layer structure in its repeated module, which interacts with each other to affect the output, such as Figure 2As shown in the figure, the structure of LSTM in the present invention includes two hidden layers and one fully connected layer (with tanh activation function). The first hidden layer receives input data and processes it through its gating units (including input gate, forget gate, output gate) and cell state. This hidden layer extracts time series features from the input data and outputs them to the next hidden layer; the second hidden layer receives the output from the first hidden layer and continues to extract more abstract time series features. After processing by this layer, the output data will be passed to the subsequent fully connected layer. The softmax layer is used for classification tasks, mapping the output of the fully connected layer to a probability distribution. The output of this layer is the probability value of each category, and these probabilities add up to 1.

[0049] Step 3: Use the power classification dataset EMBED to evaluate the performance of the LSTM found: During the algorithm operation, the slave process constructs the corresponding LSTM according to the hyperparameter combination sent by the master process, and uses the training set of the power classification dataset EMBED for training, and the test set to evaluate the current LSTM performance. After the model is trained, it is predicted on the test set and its RMSE is calculated, and the RMSE is returned to the master process as the fitness value.

[0050] The Electricity Classification Dataset EMBED is a fully labeled electricity classification dataset that contains electricity data from three apartment units in Los Angeles, California, collected for at least two weeks. In addition to the ground equipment plug load (GT) of various devices, the dataset also contains two fully labeled equipment event timestamps, making it applicable to both types of methods. Before using the Electricity Classification Dataset EMBED, it is necessary to preprocess the data, which includes the following steps: Clean the Electricity Classification Dataset EMBED to handle missing values ​​and outliers. In addition, the dataset is deduplicated to ensure the uniqueness and consistency of the data. The cleaned data is normalized, feature engineered, and converted to prepare for model training and analysis. Through data preprocessing, the data has better interpretability and applicability to improve the training effect and prediction performance of the model. The Electricity Classification Dataset EMBED is then divided into a training set and a test set according to the timestamp, providing a reliable data foundation for subsequent analysis.

[0051] The data reflects the changes in the operating status of electrical appliances in the power consumption time series. In the power classification dataset EMBED, these events have been fully labeled. The labels in the power classification dataset EMBED group events according to the device type and the operating status of the device type. The label is five digits, where the first three digits indicate the type of device and the last two digits indicate the operating mode. For example, 11101, 11102, 11103, and 11104 represent the on and off events of the refrigerator compressor, respectively.

[0052] Conduct a non-intrusive load monitoring experiment: In the experiment, the three residential units are classified using LSTM with a softmax layer. The experimental model has only one LSTM hidden layer, and the activation function of DENSE is changed to softmax, and the corresponding input and output are modified. Write a callback function for the model and observe the changes in the accuracy of model training.

[0053] The experiment labels the data in advance, conducts supervised learning, and then uses the trained model to perform classification tests on the test set. The loss on the model test set is used as the object of algorithm optimization to find the best parameter configuration for the classification model. Predictions are made on the test set and its RMSE is calculated, and the RMSE is returned to the main process as the fitness value. The algorithm ends after reaching the maximum number of iterations. The global optimal solution at this time represents the algorithm finding the best configuration of LSTM, outputting its hyperparameter configuration, and using the power classification dataset EMBED to evaluate the performance of the LSTM found.

[0054] Step 4: Update the individual optimal solution and global optimal solution of the particle: After the master process receives the fitness value returned by the slave process, it updates the individual optimal solution of the particle and the current global optimal solution of the entire particle swarm. Next, the particle weight ω is calculated according to formulas (1), (2) and (3), and then the obtained weight is used to update the speed and position, entering a new round of iteration.

[0055] The weight of a particle is calculated as follows: At each iteration, the algorithm evaluates the fitness of each particle i and obtains the fitness of all particles , , ..., , where n is the total number of particles, and then compare all , determine the maximum fitness value and the minimum fitness value respectively and , calculate the weight of the i-th particle The formula is:

[0056] (1)

[0057] (2)

[0058] (3)

[0059] in, and Represent the maximum weight value and the minimum weight value respectively. Set to the initial inertia weight during the search process, usually a larger value is taken so that the particles can search more widely at the beginning. Set to the final inertia weight in the search process, usually a smaller value is taken so that the particles are more concentrated near the current optimal solution during the later convergence. and Represents the maximum and current value of the number of iterations, is an intermediate variable used for calculation, Represents a random number that follows a uniform distribution, ranging from [0,1]. Each particle i will have an independent random number to introduce randomness, increase the diversity of particles, and prevent falling into the local optimum.

[0060] Step 5. DPSO-LSTM is an iterative algorithm that repeatedly updates the optimal solution for each individual in the group and the current local until the termination condition of the algorithm is met. DPSO-LSTM sets a maximum number of iterations. When the DPSO-LSTM algorithm reaches the specified number of iterations, it terminates. The global optimal solution at this time represents the algorithm finding the optimal parameter configuration of the constructed LSTM. The real-time power classification data set EMBED is input into the LSTM with the optimal parameter configuration to obtain the load monitoring result. The load monitoring accuracy obtained by the optimal parameter configuration and the LSTM with the optimal parameter configuration is output.

[0061] The experimental results show that the optimal parameter configuration of DPSO-LSTM is: the learning rate is 3.67e -2 , the number of hidden layer units in the first layer is 38. Under the optimal parameter configuration found by the algorithm, it has a relatively good performance. Among them, the accuracy of classifying three types of electrical appliances, refrigerators, kettles and bread machines, is as follows: Figures 4 to 6 The X-axis is composed of Epochs trained for the on state and Epochs trained for the off state, and the Y-axis is the accuracy. The average accuracy of the three appliance classifications exceeds 0.65.

[0062] Example 2

[0063] The present invention discloses a non-intrusive load monitoring device for house energy consumption, comprising:

[0064] The particle initialization module is used to randomly initialize all particles using the dynamic particle swarm optimization algorithm to obtain a combination of particle hyperparameters;

[0065] The model generation module is used to start the iterative process and generate a long short-term memory network model based on the particle hyperparameter combination;

[0066] A root mean square error acquisition module is used to divide the acquired historical power classification data set into a training set and a test set according to timestamps, train the long short-term memory network model through the training set, and then calculate the root mean square error through the test set;

[0067] The particle update module uses the root mean square error as the fitness value to update the individual optimal solution of the particle and the current global optimal solution of the entire particle swarm. It then calculates the particle weight, updates the speed and position according to the particle weight, and enters a new round of iteration.

[0068] The model optimal parameter determination module is used to repeat the above iterative process until the real-time iteration number reaches the preset maximum iteration number, determine the current global optimal solution at this time as the optimal parameter configuration of the long short-term memory network model, and obtain the long short-term memory network model with the optimal parameters;

[0069] The load monitoring result acquisition module is used to input real-time power data into the parameter-optimal long short-term memory network model to obtain load monitoring results.

[0070] Example 3

[0071] The present invention discloses a computer-readable storage medium having a computer program / instruction stored thereon. When the computer program / instruction is executed by a processor, the steps of the non-intrusive load monitoring method for house energy consumption in embodiment 1 are implemented.

[0072] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0073] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0074] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0075] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A non-intrusive load monitoring method for house energy consumption, characterized in that: The following steps are involved: Use the dynamic particle swarm optimization algorithm to randomly initialize all particles and obtain the particle hyperparameter combination; Start the iterative process to generate a LSTM network model based on the particle hyperparameter combination; The acquired historical electricity classification data set of houses is divided into a training set and a test set according to the timestamp. The long short-term memory network model is trained with the training set, and the root mean square error is calculated with the test set. The root mean square error is used as the fitness value to update the individual optimal solution of the particle and the current global optimal solution of the entire particle swarm. Then the weight of the particle is calculated, and the speed and position are updated according to the weight of the particle to enter a new round of iteration. Repeat the above iterative process until the real-time iteration number reaches the preset maximum iteration number, determine the current global optimal solution at this time as the optimal parameter configuration of the long short-term memory network model, and obtain the long short-term memory network model with the optimal parameters; The real-time power data of the house is input into the parameter-optimal long short-term memory network model to obtain the load monitoring results.

2. The non-intrusive load monitoring method for house energy consumption according to claim 1, characterized in that: The dynamic particle swarm optimization algorithm uses mixed variable encoding. Each particle represents a hyperparameter combination of the long short-term memory network. Each hyperparameter is designed with a corresponding variable. Each particle has four adjustable hyperparameter variables and their corresponding speeds.

3. The non-intrusive load monitoring method for house energy consumption according to claim 1, characterized in that: The long short-term memory network model includes two hidden layers, a fully connected layer and a softmax layer. The first hidden layer receives input data, extracts timing features from the input data through its gating units and cell states, and outputs them to the next hidden layer; the second hidden layer receives the output from the first hidden layer, and continues to extract more abstract timing features, and then passes the timing features to the fully connected layer. The fully connected layer outputs data to the softmax layer. The softmax layer is used for classification tasks and maps the output of the fully connected layer to a probability distribution.

4. The non-intrusive load monitoring method for house energy consumption according to claim 1, characterized in that: The historical power classification dataset includes ground device plug loads for various devices, and two fully labeled device event timestamps.

5. The non-intrusive load monitoring method for house energy consumption according to claim 4, characterized in that: The historical electricity classification dataset was cleaned, deduplicated, normalized, feature engineered, and transformed before use.

6. The non-intrusive load monitoring method for house energy consumption according to claim 4, characterized in that: The labels in the historical power classification dataset and real-time power data are five digits, where the first three digits indicate the type of equipment and the last two digits indicate the operation mode.

7. The non-intrusive load monitoring method for house energy consumption according to claim 1, characterized in that: The weight of a particle is calculated as follows: At each iteration, the algorithm evaluates the fitness value of each particle i and obtains the fitness value of all particles , , ..., , where n is the total number of particles, and then compare all , determine the maximum fitness value and the minimum fitness value respectively and , calculate the weight of the i-th particle The formula is: （1） （2） （3） in, and Represent the maximum weight value and the minimum weight value respectively. and Represents the maximum and current value of the number of iterations, is an intermediate variable used for calculation, Represents a random number that obeys a uniform distribution, ranging from [0,1]. Each particle i will have an independent random number.

8. The non-intrusive load monitoring method for house energy consumption according to claim 1, characterized in that: A master process and n slave processes are created through the distributed computing framework, where n is the number of particles. The particle hyperparameter combination is sent from the master process to the slave process through the remote function of the distributed computing framework, and the root mean square error is returned from the slave process to the master process.

9. A non-intrusive load monitoring device for house energy consumption, characterized in that: include; The particle initialization module is used to randomly initialize all particles using the dynamic particle swarm optimization algorithm to obtain a combination of particle hyperparameters; The model generation module is used to start the iterative process and generate a long short-term memory network model based on the particle hyperparameter combination; A root mean square error acquisition module is used to divide the acquired historical power classification data set into a training set and a test set according to timestamps, train the long short-term memory network model through the training set, and then calculate the root mean square error through the test set; The particle update module uses the root mean square error as the fitness value to update the individual optimal solution of the particle and the current global optimal solution of the entire particle swarm. It then calculates the particle weight, updates the speed and position according to the particle weight, and enters a new round of iteration. The model optimal parameter determination module is used to repeat the above iterative process until the real-time iteration number reaches the preset maximum iteration number, determine the current global optimal solution at this time as the optimal parameter configuration of the long short-term memory network model, and obtain the long short-term memory network model with the optimal parameters; The load monitoring result acquisition module is used to input real-time power data into the parameter-optimal long short-term memory network model to obtain load monitoring results.

10. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instruction is executed by a processor, the steps of the non-intrusive load monitoring method for house energy consumption described in any one of claims 1-8 are implemented.

Images