A machine learning based energy consumption monitoring method and system

Abstract

The application discloses an energy consumption monitoring method and system based on machine learning, and belongs to the technical field of data processing. The method can effectively avoid the problem of inaccurate identification caused by single data by acquiring multi-dimensional energy consumption related data corresponding to a to-be-monitored target. Machine learning training data is constructed according to the multi-dimensional energy consumption related data. The energy consumption prediction model is trained by using SLSPO based on the machine learning training data, which can effectively improve the identification accuracy of the energy consumption prediction model for data. Then, the trained energy consumption prediction model is used to identify the multi-dimensional energy consumption related data to be identified, and a future period energy consumption prediction value is obtained. Finally, energy consumption anomaly identification of the to-be-monitored target is performed according to the future period energy consumption prediction value, and energy consumption anomaly early warning is performed according to an energy consumption anomaly identification result obtained by energy consumption anomaly identification, thereby effectively improving the energy consumption anomaly monitoring accuracy and providing a strong reference for staff to implement energy-saving measures.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of data processing, and particularly relates to an energy consumption monitoring method and system based on machine learning. BACKGROUND

[0002] Energy consumption monitoring refers to a process of monitoring, recording and analyzing the energy consumption of buildings, equipment or systems in real time through a series of technical means. It aims to reveal the energy use rules and find energy waste links to provide data support for energy saving and emission reduction. Energy consumption monitoring technology includes hardware facilities (such as sensors, data collectors) and software systems (such as data processing and analysis platforms). Through these technologies, enterprises, government agencies and families can effectively understand the energy consumption status, optimize energy management, reduce operating costs and promote green and sustainable development. With the rapid development of China's economy, energy consumption problems are becoming increasingly serious. In order to improve energy utilization efficiency and reduce energy consumption, it is necessary to monitor and analyze the energy consumption of buildings or equipment in real time. Traditional energy consumption monitoring methods often directly use machine learning algorithms to identify single data, resulting in low prediction accuracy and difficulty in guiding actual energy-saving measures. SUMMARY

[0003] The application provides an energy consumption monitoring method and system based on machine learning, which solves the problem that traditional energy consumption monitoring methods often directly use machine learning algorithms to identify single data, resulting in low prediction accuracy and difficulty in guiding actual energy-saving measures.

[0004] In one aspect, the application provides an energy consumption monitoring method based on machine learning, comprising:

[0005] Obtaining multi-dimensional energy consumption related data corresponding to a to-be-monitored target, and constructing machine learning training data according to the multi-dimensional energy consumption related data;

[0006] Constructing an energy consumption prediction model using a machine learning model, and training the energy consumption prediction model using SLSPO based on the machine learning training data to obtain a trained energy consumption prediction model;

[0007] Real-time collection of to-be-identified multi-dimensional energy consumption related data, and identification of the to-be-identified multi-dimensional energy consumption related data using the trained energy consumption prediction model to obtain a future period energy consumption prediction value;

[0008] Energy consumption anomaly identification of the to-be-monitored target according to the future period energy consumption prediction value to obtain an energy consumption anomaly identification result, and energy consumption anomaly early warning according to the energy consumption anomaly identification result to complete energy consumption monitoring.

[0009] In one possible implementation, the method further comprises:

[0010] When the energy consumption anomaly recognition result is an energy consumption anomaly category, an energy saving optimization suggestion is generated, and the energy saving optimization suggestion is transmitted to a communication device designated by a staff.

[0011] In a possible implementation, multi-dimensional energy consumption related data corresponding to the target to be monitored is acquired, and machine learning training data is constructed according to the multi-dimensional energy consumption related data, including:

[0012] On the basis of a preset data sampling frequency, the energy consumption, environmental data and equipment operation data corresponding to the target to be monitored are collected by sensors in the Internet of Things to obtain multi-dimensional energy consumption related data corresponding to the target to be monitored;

[0013] The multi-dimensional energy consumption related data corresponding to the target to be monitored is preprocessed to obtain preprocessed multi-dimensional energy consumption related data;

[0014] The multi-dimensional energy consumption related data at M continuous time points is intercepted as sample multi-dimensional energy consumption related data in the form of a sliding window, and the energy consumption at the M+1 time point is extracted as an expected output label;

[0015] The sample multi-dimensional energy consumption related data and the corresponding expected output label are collectively used as machine learning training data, and multiple different machine learning training data are repeatedly acquired.

[0016] In a possible implementation, the multi-dimensional energy consumption related data corresponding to the target to be monitored is preprocessed to obtain preprocessed multi-dimensional energy consumption related data, including:

[0017] The multi-dimensional energy consumption related data corresponding to the target to be monitored is sequentially cleaned, denoised and normalized to obtain preprocessed multi-dimensional energy consumption related data.

[0018] In a possible implementation, a machine learning model is used to construct an energy consumption prediction model, and the energy consumption prediction model is trained by SLSPO on the basis of the machine learning training data to obtain a trained energy consumption prediction model, including:

[0019] A machine learning model is used to construct an energy consumption prediction model, and a particle swarm is initialized on the basis of the hyperparameters of the energy consumption prediction model.

[0020] On the basis of the machine learning training data, the fitness of each particle in the particle swarm is determined, and the optimal particle and the elite particle subgroup are determined according to the fitness of all particles.

[0021] On the basis of a preset maximum number of training times, the training process is divided into a first stage, a second stage and a third stage in sequence.

[0022] In the first stage, each particle is explored based on the elite particle subgroup using a multi-direction exploration mechanism to obtain a particle after one exploration, and the particle after one exploration is used as the initial particle of the second stage;

[0023] In the second stage, each particle is explored based on the optimal particle and the elite particle subgroup using a global and local balanced exploration mechanism to obtain a particle after two explorations, and the particle after two explorations is used as the initial particle of the third stage;

[0024] In the third stage, each particle is explored based on the elite particle subgroup using a high-precision exploration mechanism with long jumps to obtain a particle after three explorations.

[0025] Based on the particle after three explorations, the final hyperparameters corresponding to the energy consumption prediction model are determined to obtain the trained energy consumption prediction model.

[0026] In a possible implementation, after obtaining the trained energy consumption prediction model, the method further includes:

[0027] The trained energy consumption prediction model is subjected to model accuracy verification to obtain a model accuracy verification result.

[0028] It is judged whether the model accuracy verification result meets a preset accuracy requirement, if yes, no operation is performed, and the trained energy consumption prediction model is used for subsequent data processing, otherwise, the trained energy consumption prediction model is subjected to secondary training using the SLSPO or gradient descent algorithm.

[0029] In a possible implementation, in the first stage, the second stage and the third stage, after any one exploration of the particle, the particle is subjected to boundary processing.

[0030] In a possible implementation, the future period energy consumption prediction value is used for energy consumption anomaly identification of the to-be-monitored target to obtain an energy consumption anomaly identification result, including:

[0031] An actual energy consumption value corresponding to the future period is obtained, and it is judged whether a difference between the actual energy consumption value and the future period energy consumption prediction value exceeds a preset difference threshold, if yes, the energy consumption anomaly identification result is determined as energy consumption anomaly, otherwise, the energy consumption anomaly identification result is determined as normal energy consumption.

[0032] In a possible implementation, an energy consumption anomaly warning is performed according to the energy consumption anomaly identification result, including:

[0033] When the energy consumption anomaly recognition result is energy consumption anomaly, then a text warning information and / or a sound and light warning information are generated, and energy consumption anomaly warning is performed according to the text warning information and / or the sound and light warning information.

[0034] In another aspect, the present application provides an energy consumption monitoring system based on machine learning, comprising: a training data acquisition module, a machine learning module, an energy consumption prediction module, and an energy consumption comprehensive monitoring module.

[0035] The training data acquisition module is configured to acquire multi-dimensional energy consumption related data corresponding to a target to be monitored, and construct machine learning training data according to the multi-dimensional energy consumption related data.

[0036] The machine learning module is configured to construct an energy consumption prediction model using a machine learning model, and train the energy consumption prediction model using SLSPO based on the machine learning training data, to obtain a trained energy consumption prediction model.

[0037] The energy consumption prediction module is configured to collect real-time multi-dimensional energy consumption related data to be recognized, and recognize the multi-dimensional energy consumption related data to be recognized using the trained energy consumption prediction model, to obtain a future period energy consumption prediction value.

[0038] The energy consumption comprehensive monitoring module is configured to perform energy consumption anomaly recognition of the target to be monitored according to the future period energy consumption prediction value, to obtain an energy consumption anomaly recognition result, and perform energy consumption anomaly warning according to the energy consumption anomaly recognition result, to complete energy consumption monitoring.

[0039] The energy consumption monitoring method and system based on machine learning provided by the present application can effectively avoid the problem of inaccurate recognition caused by single data by acquiring multi-dimensional energy consumption related data corresponding to a target to be monitored, and construct machine learning training data according to the multi-dimensional energy consumption related data. The recognition accuracy of the energy consumption prediction model for data can be effectively improved by training the energy consumption prediction model using SLSPO based on the machine learning training data. The future period energy consumption prediction value is obtained by recognizing the multi-dimensional energy consumption related data to be recognized using the trained energy consumption prediction model. The energy consumption anomaly recognition of the target to be monitored is performed according to the future period energy consumption prediction value. The energy consumption anomaly warning is performed according to the energy consumption anomaly recognition result obtained by the energy consumption anomaly recognition, which effectively improves the accuracy of energy consumption anomaly monitoring and provides a strong reference for staff to implement energy-saving measures. BRIEF DESCRIPTION OF DRAWINGS

[0040] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0041] Figure 1This is a flowchart illustrating a machine learning-based energy consumption monitoring method provided in an embodiment of the present invention.

[0042] Figure 2 This is a schematic diagram of the structure of an energy consumption monitoring system based on machine learning, provided in an embodiment of the present invention.

[0043] The accompanying drawings have illustrated specific embodiments of the invention, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0044] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0045] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0046] like Figure 1 As shown, this embodiment of the invention provides an energy consumption monitoring method based on machine learning, including:

[0047] S101. Obtain multi-dimensional energy consumption related data corresponding to the target to be monitored, and construct machine learning training data based on the multi-dimensional energy consumption related data;

[0048] Multidimensional energy consumption data can include various types of data related to energy consumption, and should at least include some operational data, so as to more comprehensively reflect the energy consumption pattern and avoid errors caused by processing single data.

[0049] In this embodiment of the invention, the target to be monitored can be a building, thereby enabling the monitoring of the building's energy consumption.

[0050] S102. Construct an energy consumption prediction model using a machine learning model, and train the energy consumption prediction model using SLSPO (Stage learning Particle Swarm Optimization) based on the machine learning training data to obtain the trained energy consumption prediction model.

[0051] The machine learning model can be used to build the energy consumption prediction model, which can include using machine learning models such as convolutional neural networks, bidirectional long short-term memory networks, BP neural networks, recurrent neural networks, etc. to build the energy consumption prediction model. However, it should be noted that other machine learning models (such as support vector machines) can also be used to build the energy consumption prediction model. The preferred embodiment of the present application uses a convolutional neural network to process multi-dimensional data to achieve energy consumption prediction.

[0052] The prior art generally uses a particle swarm algorithm to optimize the hyperparameters of the machine learning model, but it is prone to local optimization, which reduces the energy consumption prediction ability. Therefore, the present application proposes a stage learning particle swarm algorithm to improve the global search ability and optimize the hyperparameters.

[0053] S103, real-time collection of multi-dimensional energy consumption related data to be identified, and identification of the multi-dimensional energy consumption related data to be identified using the trained energy consumption prediction model to obtain future period energy consumption prediction values;

[0054] The trained energy consumption prediction model has the ability to predict energy consumption, so the multi-dimensional energy consumption related data to be identified can be used as input to the trained energy consumption prediction model to obtain future period energy consumption prediction values. It should be noted that the data form and type of the multi-dimensional energy consumption related data to be identified should be the same as the input data in the machine learning training data, so that the trained energy consumption prediction model can correctly identify the data.

[0055] S104, energy consumption anomaly identification of the monitored target according to the future period energy consumption prediction values to obtain energy consumption anomaly identification results, and energy consumption anomaly warning according to the energy consumption anomaly identification results to complete energy consumption monitoring.

[0056] When the future period energy consumption prediction values are identified, the subsequent real data can be monitored to determine the energy consumption anomaly identification results. When the energy consumption anomaly identification results are energy consumption anomaly categories, the warning can be realized to enable the staff to optimize energy consumption.

[0057] In one possible implementation, the method further comprises:

[0058] When the energy consumption anomaly identification result is an energy consumption anomaly category, an energy saving optimization suggestion is generated and transmitted to the communication device designated by the staff.

[0059] For example, if the prediction result shows that the energy consumption is high at a certain period, it is recommended to adjust the operation time of non-essential equipment to a period with lower energy consumption. For example, the operation time of equipment such as air conditioning, lighting, etc. is avoided during the peak period of electricity consumption. For industrial equipment, it is recommended to optimize the production plan and arrange the operation time of high energy consumption equipment in the period with lower electricity price. According to the prediction result, the environmental parameters are adjusted to reduce energy consumption. For example, when high temperature weather is predicted, the air conditioning temperature setting value is adjusted in advance to avoid excessive refrigeration. For lighting systems, it is recommended to adjust the light brightness according to the light prediction result to make full use of natural light. If the prediction result shows that the equipment load is too high, it is recommended to optimize the equipment load distribution to avoid overloading of a single device. For example, in a data center, the computing task is distributed to multiple servers to reduce the energy consumption of a single server.

[0060] It is worth noting that the above energy-saving optimization suggestions are only examples, and the corresponding energy-saving optimization suggestions can be set for each abnormal energy consumption category by the staff according to actual needs. The abnormal energy consumption category can be set as one category of abnormal energy consumption, or as a specific small category, such as abnormal energy consumption, etc.

[0061] In one possible implementation, the multi-dimensional energy consumption related data corresponding to the to-be-monitored target is obtained, and machine learning training data is constructed according to the multi-dimensional energy consumption related data, including:

[0062] The multi-dimensional energy consumption related data corresponding to the to-be-monitored target is obtained by collecting the energy consumption, environmental data and equipment operation data corresponding to the to-be-monitored target through sensors in the Internet of Things based on a preset data sampling frequency;

[0063] The environmental data can be temperature, illumination, humidity, air quality, etc. related to the environment, and holiday information and workday information can also be added to improve the prediction accuracy of building energy consumption. The equipment operation data can include the temperature, voltage, current, switch state of the equipment, and the fault alarm information of the equipment collected through the equipment monitoring system, etc. For motor type equipment, the rotating speed of the motor can also be collected. However, it is worth noting that the above data types are only examples, and other data can also be used as multi-dimensional energy consumption related data.

[0064] The multi-dimensional energy consumption related data corresponding to the to-be-monitored target is preprocessed to obtain preprocessed multi-dimensional energy consumption related data;

[0065] The multi-dimensional energy consumption related data at M consecutive time points is intercepted as sample multi-dimensional energy consumption related data in the form of a sliding window, and the energy consumption at the M+1 time point is extracted as the expected output label;

[0066] The sample multi-dimensional energy consumption related data and the corresponding expected output label are collectively used as machine learning training data, and multiple different machine learning training data are repeatedly obtained.

[0067] In a possible implementation, the multi-dimensional energy consumption related data corresponding to the target to be monitored is preprocessed to obtain the multi-dimensional energy consumption related data after preprocessing, including:

[0068] The multi-dimensional energy consumption related data corresponding to the target to be monitored is sequentially cleaned, denoised, and normalized to obtain the multi-dimensional energy consumption related data after preprocessing.

[0069] The cleaning can be processing of abnormal values, the denoising can be filtering of data, and the normalization processing reduces the complexity of data, thereby improving the ability to identify data. However, it should be noted that in the process of training the machine learning model, if the multi-dimensional energy consumption related data is directly input by the staff, it can also not be processed, and the data identification can also be achieved.

[0070] The data is normalized here, and in order to ensure the accurate identification of subsequent data, the data input into the energy consumption prediction model also needs to be normalized.

[0071] In a possible implementation, the machine learning model is used to construct the energy consumption prediction model, and the SLSPO is used to train the energy consumption prediction model based on the machine learning training data to obtain the trained energy consumption prediction model, including:

[0072] A1, the machine learning model is used to construct the energy consumption prediction model, and the particle swarm is initialized based on the hyperparameters of the energy consumption prediction model;

[0073] A2, based on the machine learning training data, the fitness of each particle in the particle swarm is determined, and the optimal particle and the elite particle subpopulation are determined according to the fitness of all particles;

[0074] For example, the K particles with the largest fitness can be selected to form the elite particle subpopulation, and K should be less than N / 3, N representing the total number of particles in the particle swarm.

[0075] A3, based on the preset maximum training number, the training process is divided into the first stage, the second stage, and the third stage which are sequentially performed;

[0076] The first stage can be a stage where t < T / 3, the second stage can be a stage where T / 3 ≤ t < 2T / 3, and the third stage can be a stage where 2T / 3 ≤ t. The training number t gradually increases from 1 starting from the first training, so when the training number t accumulates to a certain value, the next stage is entered, and the training is completed until the training is completed.

[0077] A4. In the first stage, based on the elite particle subgroup, a multi-directional exploration mechanism is used to explore each particle to obtain the particle after one exploration, and the particle after one exploration is used as the initial particle for the second stage.

[0078] A4.1 For any given particle, randomly select a reference particle from the elite particle subgroup to obtain the reference particle corresponding to the particle.

[0079] A4.2. Based on the reference particle corresponding to the aforementioned particle, the update rate corresponding to the particle is obtained as follows:

[0080]

[0081] in, Indicates the first t During the training process, the first n There are 1, 2, ..., N particles, where N represents the total number of particles in the swarm. Indicates the first t During the training process, the first n The update rate corresponding to each particle Indicates the first t +1 training session n The update rate corresponding to each particle Indicates inertia weight, Indicates the first t During the training process, the first n The reference particle corresponding to each particle Represents a random number that follows a Brownian distribution;

[0082] A4.3. Based on the update rate corresponding to the particle, explore the particle to obtain the particle after one exploration:

[0083]

[0084] in, This represents the particles after an exploration. This represents a constant term, and is set to 0.5. It represents a random number that follows a uniform distribution between [0,1].

[0085] It is worth noting that the elite particle swarm needs to be updated in each exploration during the first phase, and the same applies to the subsequent second and third phases, which will not be elaborated upon further.

[0086] The multi-direction exploration mechanism provided by the embodiment of the present application can make the whole particle group explore different better directions in the solution space, and the exploration speed is relatively fast, so that the global search can be effectively realized, the global search capability of the algorithm is improved, and the optimal region in the solution space can be quickly found.

[0087] A5. In the second stage, based on the optimal particle and the elite particle subgroup, a global and local balanced exploration mechanism is used to explore each particle, to obtain the particles after secondary exploration, and the particles after secondary exploration are taken as initial particles in the third stage.

[0088] A5.1. The fitness corresponding to the particles after primary exploration is obtained, and the particles with greater fitness are taken as local exploration particles, and the particles with smaller fitness are taken as global exploration particles.

[0089] A5.2. The local exploration particles are updated according to the optimal particle, and the updated local exploration particles are obtained as follows:

[0090]

[0091]

[0092] wherein, represents the i-th local exploration particle in the n-th training process, t =1, 2, …, N / 2, N is an even number, i represents the optimal particle, i represents the preset maximum training number, represents the circular constant, represents a random number between (0, 1), represents a cosine function, represents the i-th local exploration particle in the n+1-th training process, represents the updating speed of the i-th local exploration particle in the n+1-th training process, represents the updated local exploration particle; t i A5.3. The global exploration particles are updated according to the elite particle subgroup, and the updated global exploration particles are obtained as follows:

[0093]

[0094]

[0095]

[0096]

[0097] wherein, represents the i-th global exploration particle in the n-th training process, t ​​​the i-th global exploration particle in the i-th training process, j j i=1,2,…,N / 2, denotes the global exploration particle in the elite particle subgroup randomly matched reference particle, denotes the i-th global exploration particle in the i-th training process, t the i-th global exploration particle in the i-th training process, j denotes the update speed of the i-th global exploration particle in the i-th training process, denotes the update speed of the i-th global exploration particle in the i-th training process, t the i-th global exploration particle in the i-th training process, j the i-th global exploration particle in the i-th training process, denotes the global exploration particle after update, denotes the global search control parameter;

[0098] A5.4, the local exploration particle after update and the global exploration particle after update are collectively used as the particle after secondary exploration.

[0099] The global and local balanced exploration mechanism provided by the embodiment of the application can make part of the particles with better positions in the solution space perform local search, ensure the convergence ability of the algorithm, and make part of the particles with poor positions perform global search around the position of the elite particle, assist the algorithm to jump out of the local optimal solution, and ensure the ability of searching the optimal solution finally.

[0100] A6, in the third stage, based on the elite particle subgroup, a high-precision exploration mechanism with long jump is used to explore each particle, and the particle after tertiary exploration is obtained;

[0101]

[0102]

[0103] wherein, the i-th particle after secondary exploration in the i-th training process, t the i-th particle after secondary exploration in the i-th training process, k denotes the update speed of the i-th particle after secondary exploration in the i-th training process, k i=1,2,…,N, denotes the historical optimal value corresponding to the particle denotes the update speed of the i-th particle after secondary exploration in the i-th training process, denotes the reduction factor, t the i-th particle after secondary exploration in the i-th training process, k the i-th particle after secondary exploration in the i-th training process, denotes the first learning factor, the i-th particle after secondary exploration in the i-th training process, t the i-th particle after secondary exploration in the i-th training process, k the i-th particle after secondary exploration in the i-th training process, the i-th particle after secondary exploration in the i-th training process, ​This represents a random number between (0,1) generated by Levi's flight. Indicates the second learning factor. This represents a random number between (0,1) generated by Levi's flight. Represents a random number between (0,1). Indicates the first k The fitness value corresponding to each particle can be: , Indicates the first k The loss function value corresponding to each particle, where e represents the natural constant.

[0104] The high-precision exploration mechanism with long jump provided in this embodiment of the invention introduces a reduction factor and Lévy flight random numbers, which can ensure convergence accuracy while enabling particles to have a certain global exploration capability. At the same time, a large perturbation is introduced to further enhance the global optimization capability of the particle swarm, and the perturbation gradually decreases, thereby ensuring the final convergence accuracy. Combined with the reduction factor, it can effectively solve the problem that existing particle swarm algorithms are prone to getting trapped in local optima, and also ensure convergence accuracy.

[0105] A7. Based on the particles after three explorations, determine the final hyperparameters corresponding to the energy consumption prediction model, and obtain the trained energy consumption prediction model.

[0106] This invention employs SLSPO to train the energy consumption prediction model, which effectively enhances the machine learning model's ability to learn from data and improves training results. Compared to the original particle swarm optimization algorithm, it effectively improves the accuracy of data preprocessing in the trained energy consumption prediction model, ensuring the effectiveness of energy consumption monitoring.

[0107] In one possible implementation, after obtaining the trained energy consumption prediction model, the process further includes:

[0108] The accuracy of the trained energy consumption prediction model is verified to obtain the model accuracy verification results.

[0109] Determine whether the model accuracy verification result meets the preset accuracy requirements. If yes, no operation is performed, and the trained energy consumption prediction model is used for subsequent data processing. Otherwise, the trained energy consumption prediction model is trained a second time using SLSPO or gradient descent algorithm.

[0110] In one possible implementation, after any one exploration of the particle in the first, second, and third phases, the process further includes: handling the particle for out-of-bounds errors.

[0111] In one possible implementation, energy consumption anomaly identification of the target to be monitored is performed based on the predicted energy consumption value for the future time period, resulting in an energy consumption anomaly identification result, including:

[0112] obtaining an actual energy consumption value corresponding to the future period, and determining whether a difference between the actual energy consumption value and the predicted energy consumption value of the future period exceeds a preset difference threshold value, if yes, determining that the energy consumption anomaly recognition result is energy consumption anomaly, otherwise, determining that the energy consumption anomaly recognition result is normal energy consumption.

[0113] A reasonable threshold value (such as ±10%) can be set, and if the deviation between the actual energy consumption and the predicted energy consumption exceeds the threshold value, it is determined to be abnormal. For example, if the predicted energy consumption is 100 kWh, the actual energy consumption is 120 kWh, the deviation is 20%, which exceeds the threshold value, and it is determined to be abnormal.

[0114] In one possible implementation, energy consumption anomaly early warning is performed according to the energy consumption anomaly recognition result, including:

[0115] When the energy consumption anomaly recognition result is energy consumption anomaly, text early warning information and / or sound and light early warning information are generated, and energy consumption anomaly early warning is performed according to the text early warning information and / or sound and light early warning information.

[0116] For example, when there is a field early warning device, sound and light alarm information can be generated, and the sound and light alarm information is transmitted to the field early warning device for energy consumption anomaly early warning. The text early warning information can be transmitted to the corresponding device of the staff for energy consumption anomaly early warning. The specific early warning method can be set according to the actual needs of the staff.

[0117] The energy consumption monitoring method based on machine learning provided by the present application can effectively avoid the problem of inaccurate recognition caused by single data by obtaining multi-dimensional energy consumption related data corresponding to the target to be monitored, and can effectively improve the recognition accuracy of the energy consumption prediction model for data by constructing machine learning training data according to the multi-dimensional energy consumption related data, training the energy consumption prediction model using SLSPO based on the machine learning training data, and then identifying the multi-dimensional energy consumption related data to be identified using the trained energy consumption prediction model to obtain the future period energy consumption prediction value. Finally, the energy consumption anomaly of the target to be monitored is recognized according to the future period energy consumption prediction value, and the energy consumption anomaly early warning is performed according to the energy consumption anomaly recognition result obtained by the energy consumption anomaly recognition, which effectively improves the energy consumption anomaly monitoring accuracy and provides a powerful reference for the staff to implement energy-saving measures.

[0118] As shown in Figure 2 The energy consumption monitoring system based on machine learning provided by the present application embodiment includes a training data acquisition module 201, a machine learning module 202, an energy consumption prediction module 203, and an energy consumption comprehensive monitoring module 204.

[0119] The training data acquisition module 201 is configured to acquire multi-dimensional energy consumption related data corresponding to the target to be monitored, and construct machine learning training data according to the multi-dimensional energy consumption related data.

[0120] The machine learning module 202 is configured to construct an energy consumption prediction model by using a machine learning model, and train the energy consumption prediction model by using SLSPO based on the machine learning training data, to obtain the trained energy consumption prediction model.

[0121] The energy consumption prediction module 203 is configured to collect multi-dimensional energy consumption related data to be identified in real time, and identify the multi-dimensional energy consumption related data to be identified by using the trained energy consumption prediction model, to obtain a future time period energy consumption prediction value.

[0122] The energy consumption comprehensive monitoring module 204 is configured to perform energy consumption anomaly identification of the target to be monitored according to the future time period energy consumption prediction value, to obtain an energy consumption anomaly identification result, and perform energy consumption anomaly early warning according to the energy consumption anomaly identification result, to complete energy consumption monitoring.

[0123] Figure 2 The machine learning-based energy consumption monitoring system can execute the method technical solution, and has similar principles and beneficial effects, which will not be described here.

[0124] Those skilled in the art should understand that embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can be in 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.

[0125] The present application is described with reference to flowcharts and / or block diagrams according to the method, device (system), and computer program product of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one flow or multiple flows and / or blocks Figure 1 The device that implements the functions specified in one block or multiple blocks.

[0126] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0127] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0128] Those of ordinary skill in the art can understand that all or part of the steps of the above-mentioned facts and methods can be completed by programs instructing relevant hardware, and the programs involved or the programs mentioned can be stored in a computer-readable storage medium. When the program is executed, it includes the following steps: at this time, the corresponding method steps are derived, and the storage medium can be ROM / RAM, a magnetic disc, an optical disc, etc.

[0129] The above detailed description of the specific implementation of the present application further explains the purpose, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific implementation of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A machine learning based energy consumption monitoring method, characterized by, The method comprises the following steps: acquiring multi-dimensional energy consumption related data corresponding to the target to be monitored, and constructing machine learning training data according to the multi-dimensional energy consumption related data; constructing an energy consumption prediction model by using a machine learning model, and training the energy consumption prediction model by using a stage learning particle swarm optimization (SLSPO) algorithm based on the machine learning training data to obtain a trained energy consumption prediction model; real-time acquisition of multi-dimensional energy consumption related data to be identified, and identification of the multi-dimensional energy consumption related data to be identified by using the trained energy consumption prediction model to obtain a future time period energy consumption prediction value; energy consumption anomaly identification of the target to be monitored based on the future time period energy consumption prediction value to obtain an energy consumption anomaly identification result, and energy consumption anomaly early warning based on the energy consumption anomaly identification result to complete energy consumption monitoring; constructing an energy consumption prediction model by using a machine learning model, and training the energy consumption prediction model by using a SLSPO algorithm based on the machine learning training data to obtain a trained energy consumption prediction model, which comprises the following steps: constructing an energy consumption prediction model by using a machine learning model, and initializing a particle swarm based on hyperparameters of the energy consumption prediction model; determining the fitness of each particle in the particle swarm based on the machine learning training data, and determining an optimal particle and an elite particle sub-swarm based on the fitness of all particles; dividing the training process into a first stage, a second stage and a third stage in sequence based on a preset maximum training number; in the first stage, exploring each particle by using a multi-directional exploration mechanism based on the elite particle sub-swarm to obtain particles after one-time exploration, and taking the particles after one-time exploration as initial particles of the second stage; in the second stage, exploring each particle by using a global and local balanced exploration mechanism based on the optimal particle and the elite particle sub-swarm to obtain particles after two-time exploration, which comprises the following steps: acquiring the fitness of the particles after one-time exploration, and taking particles with larger fitness as local exploration particles and particles with smaller fitness as global exploration particles; updating the local exploration particles based on the optimal particle to obtain updated local exploration particles; wherein, denotes the i-th local exploration particle in the j-th training process, t i denotes the i-th local exploration particle in the j-th training process, i =1, 2, …, N / 2, N is even, denotes the optimal particle, denotes the preset maximum training number, denotes the pi, denotes the random number between (0, 1), denotes the cosine function, denotes the i-th local exploration particle in the j-th training process, t denotes the i-th local exploration particle in the j-th training process, i denotes the i-th local exploration particle in the j-th training process, denotes the updated local exploration particle;​ updating the global exploration particles based on the elite particle sub-swarm to obtain updated global exploration particles; wherein, denotes the i-th global exploration particle in the j-th training process, t denotes the i-th global exploration particle in the j-th training process, j denotes the i-th global exploration particle in the j-th training process, j = 1, 2, …, N / 2, denotes the i-th global exploration particle in the j-th training process, denotes the i-th global exploration particle in the j-th training process, denotes the i-th global exploration particle in the j-th training process, t denotes the i-th global exploration particle in the j-th training process, j denotes the i-th global exploration particle in the j-th training process, denotes the i-th global exploration particle in the j-th training process, t denotes the i-th global exploration particle in the j-th training process, j denotes the i-th global exploration particle in the j-th training process, denotes the i-th global exploration particle in the j-th training process, denotes the i-th global exploration particle in the j-th training process, taking the updated local exploration particles and the updated global exploration particles together as the particles after two-time exploration; taking the particles after two-time exploration as initial particles of the third stage; in the third stage, exploring each particle by using a high-precision exploration mechanism with long jump based on the elite particle sub-swarm to obtain particles after three-time exploration; determining final hyperparameters of the energy consumption prediction model based on the particles after three-time exploration to obtain the trained energy consumption prediction model. 2.The machine learning based energy consumption monitoring method of claim 1, wherein, Further comprising: when the energy consumption anomaly identification result is an energy consumption anomaly category, generating an energy saving optimization suggestion, and transmitting the energy saving optimization suggestion to a communication device designated by a staff member. 3.The machine learning based energy consumption monitoring method of claim 1, wherein, acquiring multi-dimensional energy consumption related data corresponding to the target to be monitored, and constructing machine learning training data according to the multi-dimensional energy consumption related data, which comprises the following steps: The sensor in the Internet of Things collects energy consumption, environmental data and equipment operation data of the target to be monitored at a preset data sampling frequency, to obtain multi-dimensional energy consumption related data corresponding to the target to be monitored; The multi-dimensional energy consumption related data corresponding to the target to be monitored is preprocessed to obtain preprocessed multi-dimensional energy consumption related data; The multi-dimensional energy consumption related data at M consecutive time points is intercepted as sample multi-dimensional energy consumption related data in the form of a sliding window, and the energy consumption at the M+1 time point is extracted as an expected output label; The sample multi-dimensional energy consumption related data and the corresponding expected output label are used as machine learning training data, and multiple different machine learning training data are repeatedly obtained. 4.The machine learning based energy consumption monitoring method of claim 3, wherein, The multi-dimensional energy consumption related data corresponding to the target to be monitored is preprocessed to obtain preprocessed multi-dimensional energy consumption related data, including: The multi-dimensional energy consumption related data corresponding to the target to be monitored is sequentially cleaned, denoised and normalized to obtain preprocessed multi-dimensional energy consumption related data. 5.The machine learning based energy consumption monitoring method of claim 4, wherein, After obtaining the trained energy consumption prediction model, it further includes: The model accuracy of the trained energy consumption prediction model is verified to obtain a model accuracy verification result; If the model accuracy verification result meets the preset accuracy requirement, no operation is performed, and the trained energy consumption prediction model is used for subsequent data processing; otherwise, the trained energy consumption prediction model is retrained using the SLSPO or gradient descent algorithm. 6.The machine learning based energy consumption monitoring method of claim 5, wherein, In the first stage, the second stage and the third stage, after any one exploration of the particle, the particle is further processed to determine whether the model accuracy verification result meets the preset accuracy requirement. 7.The machine learning based energy consumption monitoring method of claim 6, wherein, According to the future period energy consumption prediction value, the energy consumption anomaly of the target to be monitored is identified to obtain an energy consumption anomaly identification result, including: The energy consumption actual value corresponding to the future period is obtained, and it is judged whether the difference between the energy consumption actual value and the future period energy consumption prediction value exceeds the preset difference threshold. If yes, it is determined that the energy consumption anomaly identification result is energy consumption anomaly, otherwise it is determined that the energy consumption anomaly identification result is energy consumption normal. 8.The machine learning based energy consumption monitoring method of claim 1, wherein, According to the energy consumption anomaly identification result, an energy consumption anomaly warning is performed, including: When the energy consumption anomaly identification result is energy consumption anomaly, a text warning information and / or a sound and light warning information is generated, and an energy consumption anomaly warning is performed according to the text warning information and / or the sound and light warning information. 9.A machine learning based energy consumption monitoring system for performing the machine learning based energy consumption monitoring method according to any one of claims 1 to 8. It includes: A training data acquisition module, a machine learning module, an energy consumption prediction module and an energy consumption comprehensive monitoring module; The training data acquisition module is configured to acquire multi-dimensional energy consumption related data corresponding to the target to be monitored, and construct machine learning training data based on the multi-dimensional energy consumption related data; The machine learning module is configured to construct an energy consumption prediction model using a machine learning model, and train the energy consumption prediction model using SLSPO based on the machine learning training data to obtain a trained energy consumption prediction model; The energy consumption prediction module is configured to collect real-time multi-dimensional energy consumption related data to be identified, and identify the multi-dimensional energy consumption related data to be identified using the trained energy consumption prediction model to obtain a future period energy consumption prediction value; The energy consumption comprehensive monitoring module is configured to perform energy consumption anomaly identification on the to-be-monitored target according to the future time period energy consumption prediction value, obtain an energy consumption anomaly identification result, perform energy consumption anomaly early warning according to the energy consumption anomaly identification result, and complete energy consumption monitoring.

Images