Abnormal identification method and device for energy consumption structure data, storage medium and terminal
By constructing the energy consumption structure matrix and reconstructing it using the adaptive neural network model, identifying enterprise energy consumption structure data abnormalities, solving the problem of low accuracy in the existing technology, and achieving more efficient energy management and identification.
Patent Information
- Application Number
- CN202510251241.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-04
- Publication Date
- 2025-07-29
AI Technical Summary
In the prior art, the accuracy of abnormal identification of enterprise energy consumption structure data is low, which affects the economic benefits and environmental protection of enterprises.
By constructing the energy-using structure matrix, the energy-using structure matrix is reconstructed using the adaptively adjusted neural network model, the reconstruction matrix and reconstruction error threshold are generated, and the energy-using structure point exceptions and context exceptions are identified.
It improves the accuracy of abnormal identification of energy-using structure data, reduces dependence on artificial experience, reduces the interference of subjective factors in manual analysis, optimizes energy management, and improves energy utilization efficiency.
Smart Images
Figure CN120387110A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of big data analysis, and particularly to a method and device for abnormal identification of energy consumption structure data, a storage medium, and a terminal. Background Art
[0002] The energy consumption structure refers to the proportion and composition of various energies used by an enterprise or organization during the production and operation process. With the continuous growth of global energy demand and the increasing awareness of environmental protection, the rationality and efficiency of the enterprise's energy use structure have become the focus of social attention. However, in actual operation, abnormal phenomena in the enterprise's energy use structure occur from time to time, which not only affects the economic benefits of the enterprise but also may cause serious environmental pollution. At the same time, the energy consumption structure of the enterprise can also reflect the enterprise's management status from the side, which is of great significance for enterprise evaluation.
[0003] Abnormalities in the enterprise's energy use structure usually manifest as problems such as unbalanced energy consumption ratios, low energy utilization efficiency, and serious energy waste. These problems may stem from various factors, such as defects in the optimization and upgrading of the industrial structure, unreasonable energy structure design, and poor energy management. Currently, only professional personnel can evaluate based on experience, and it is impossible to accurately identify abnormalities in the enterprise's energy consumption structure data. Summary of the Invention
[0004] In view of this, the present invention provides a method and device for abnormal identification of energy consumption structure data, a storage medium, and a terminal, mainly aiming to solve the problem of low accuracy in abnormal identification of existing enterprise energy consumption structure data.
[0005] According to one aspect of the present invention, a method for abnormal identification of energy consumption structure data is provided, including:
[0006] Obtain the energy consumption data of various energies at different time nodes of the target enterprise within the historical period, and construct an energy consumption structure matrix based on the energy consumption data, where the energy consumption structure matrix includes the energy proportions of each energy type at different time nodes;
[0007] Use the trained energy consumption structure data reconstruction model to perform reconstruction processing on the energy consumption structure matrix to obtain an energy consumption structure reconstruction matrix, where the number of hidden layers and the input-output dimension parameters of the energy consumption structure data reconstruction model are determined through adaptive adjustment, and the energy consumption structure reconstruction matrix is used to represent the energy consumption structure distribution under normal circumstances;
[0008] Determine the reconstruction error and the reconstruction error threshold based on the energy consumption structure matrix and the energy consumption structure reconstruction matrix;
[0009] Generate the abnormal recognition result of the target enterprise's energy consumption structure data based on the reconstruction error and the reconstruction error threshold, including the abnormal energy consumption structure point and the abnormal energy consumption structure context.
[0010] Further, before the method uses the trained energy consumption structure data reconstruction model to reconstruct the energy consumption structure matrix to obtain the energy consumption structure reconstruction matrix, the method further includes:
[0011] Obtain the normal energy consumption structure matrix sample, and construct an encoder and a decoder respectively;
[0012] According to the normal energy consumption structure matrix sample input to the encoder, adaptively adjust the number of hidden layers and the input and output dimension parameters to obtain the initial energy consumption structure data reconstruction model;
[0013] Construct an encoder loss function and a decoder loss function respectively, and determine the difference between the encoder loss function and the decoder loss function as the joint loss function;
[0014] Use the joint loss function to train the initial energy consumption structure data reconstruction model to obtain the trained energy consumption structure data reconstruction model.
[0015] Further, the step of, according to the normal energy consumption structure matrix sample input to the encoder, adaptively adjusting the number of hidden layers and the input and output dimension parameters to obtain the initial energy consumption structure data reconstruction model includes:
[0016] According to the dimension of the normal energy consumption structure matrix sample and the preset latent variable distribution dimension, calculate the number of model hidden layers according to the hidden layer adaptive logarithmic function, wherein the dimension of the normal energy consumption structure matrix sample and the preset latent variable distribution dimension are in a logarithmic function relationship with the number of model hidden layers;
[0017] Calculate the dimension, input dimension and output dimension of each hidden layer according to the number of model hidden layers to obtain the initial energy consumption structure data reconstruction model.
[0018] Further, the construction process of the encoder loss function includes:
[0019] Use the encoder to encode the normal energy consumption structure matrix sample to obtain the posterior distribution of the latent variables of the normal energy consumption structure matrix sample;
[0020] Construct an encoder loss function based on the KL divergence between the posterior distribution of the latent variables and the standard normal distribution;
[0021] The construction process of the decoder loss function includes:
[0022] Decode the latent variables of the normal energy consumption structure matrix sample using a decoder to obtain the reconstruction result of the normal energy consumption structure matrix sample;
[0023] Construct a decoder loss function based on the expectation of the log-likelihood of the reconstruction result on the posterior distribution of the latent variables, so as to balance the difference between the normal energy consumption structure matrix sample and the reconstruction result by taking the expectation of the posterior distribution of the latent variables.
[0024] Further, the constructing the energy consumption structure matrix based on the energy consumption data includes:
[0025] Convert the energy consumption data according to the standard coal conversion coefficient corresponding to the energy type of the energy consumption data to obtain the standard coal consumption of each energy type at different time nodes;
[0026] For each energy type, calculate the proportion of the standard coal consumption of the energy type relative to the total standard coal consumption of the global energy types, and construct an energy consumption structure matrix based on the energy proportions of each energy type at different time nodes.
[0027] Further, the determining the reconstruction error and the reconstruction error threshold based on the energy consumption structure matrix and the energy consumption structure reconstruction matrix includes:
[0028] Calculate the differences between the energy proportions corresponding to the same energy type at the same time node in the energy consumption structure matrix and the energy consumption structure reconstruction matrix respectively, and sum the differences of different energy types at the same time node to obtain the reconstruction errors at different time nodes;
[0029] Sort the reconstruction error values in descending order, and generate a reconstruction error curve with the sorting order of the reconstruction error as the horizontal axis and the value of the reconstruction error as the vertical axis;
[0030] Extract the inflection point with the largest inflection slope in the reconstruction error curve, and determine the reconstruction error corresponding to the inflection point as the reconstruction error threshold.
[0031] Further, the generating the abnormal identification result of the target enterprise energy consumption structure data based on the reconstruction error and the reconstruction error threshold includes:
[0032] Determine the time nodes corresponding to the reconstruction errors greater than the reconstruction error threshold as abnormal energy consumption nodes;
[0033] If the number of the abnormal energy consumption nodes within a preset time period is greater than the abnormal node threshold, determine that the abnormal identification result of the energy consumption structure is an abnormal energy consumption structure set, where the abnormal node threshold is less than the number of time nodes included in the preset time period;
[0034] If the number of abnormal energy - using nodes is less than or equal to the abnormal node threshold, it is determined that the abnormal recognition result of the energy - using structure is an energy - using structure point abnormality;
[0035] Sort the reconstruction errors in ascending order of time nodes. If, for two adjacent reconstruction errors, the first reconstruction error at the earlier order is greater than the second reconstruction error at the later order, and the absolute value of their difference is greater than a preset multiple of the first reconstruction error, it is determined that the abnormal recognition result of the energy - using structure is an energy - using structure context abnormality.
[0036] According to another aspect of the present invention, there is provided an apparatus for identifying abnormal energy - using structure data, including:
[0037] An acquisition module, configured to acquire the energy - using data of various types of energy at different time nodes of a target enterprise within a historical period, and construct an energy - using structure matrix based on the energy - using data, where the energy - using structure matrix includes the energy proportion of each energy type at different time nodes;
[0038] A reconstruction module, configured to perform a reconstruction process on the energy - using structure matrix by using a trained energy - using structure data reconstruction model to obtain an energy - using structure reconstruction matrix, where the number of hidden layers and the input - output dimension parameters of the energy - using structure data reconstruction model are determined by adaptive adjustment, and the energy - using structure reconstruction matrix is used to represent the energy - using structure distribution under normal circumstances;
[0039] A determination module, configured to determine a reconstruction error and a reconstruction error threshold based on the energy - using structure matrix and the energy - using structure reconstruction matrix;
[0040] A generation module, configured to generate an abnormal recognition result of the energy - using structure data of the target enterprise based on the reconstruction error and the reconstruction error threshold, including an energy - using structure point abnormality and an energy - using structure context abnormality.
[0041] According to still another aspect of the present invention, there is provided a storage medium in which at least one executable instruction is stored, and the executable instruction causes a processor to perform operations corresponding to the above - mentioned method for identifying abnormal energy - using structure data.
[0042] According to yet another aspect of the present invention, there is provided a terminal, including: a processor, a memory, a communication interface, and a communication bus, and the processor, the memory, and the communication interface complete communication with each other through the communication bus;
[0043] The memory is used to store at least one executable instruction, and the executable instruction causes the processor to perform operations corresponding to the above - mentioned method for identifying abnormal energy - using structure data.
[0044] By means of the above - mentioned technical solutions, the technical solutions provided by the embodiments of the present invention have at least the following advantages:
[0045] The present invention provides a method and device for identifying anomalies in energy consumption structure data, a storage medium, and a terminal. In the embodiments of the present invention, by obtaining the energy consumption data of various types of energy at different time nodes within a historical period for a target enterprise, an energy consumption structure matrix is constructed based on the energy consumption data, where the energy consumption structure matrix includes the energy proportion of each energy type at different time nodes; the constructed energy consumption structure matrix is reconstructed by using a trained energy consumption structure data reconstruction model to obtain an energy consumption structure reconstruction matrix, where the number of hidden layers and the input-output dimension parameters of the energy consumption structure data reconstruction model are determined by adaptive adjustment, and the energy consumption structure reconstruction matrix is used to represent the energy consumption structure distribution under normal circumstances; the reconstruction error and the reconstruction error threshold are determined based on the energy consumption structure matrix and the energy consumption structure reconstruction matrix; an anomaly identification result of the energy consumption structure data of the target enterprise is generated based on the reconstruction error and the reconstruction error threshold, including energy consumption structure point anomalies and energy consumption structure context anomalies. By reconstructing the energy consumption structure matrix based on a neural network model and using the reconstruction error as the basis for judging anomalies, the dependence on manual experience is greatly reduced, and the interference of subjective factors in manual analysis is reduced, thereby greatly improving the accuracy of anomaly identification of energy consumption structure data.
[0046] The above description is only an overview of the technical solution of the present invention. In order to be able to understand the technical means of the present invention more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features, and advantages of the present invention more obvious and understandable, the following specifically describes the embodiments of the present invention. Brief Description of the Drawings
[0047] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention. And throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:
[0048] Figure 1 shows a flowchart of a method for identifying anomalies in energy consumption structure data provided by an embodiment of the present invention;
[0049] Figure 2 shows a flowchart of another method for identifying anomalies in energy consumption structure data provided by an embodiment of the present invention;
[0050] Figure 3 shows a block diagram of a device for identifying anomalies in energy consumption structure data provided by an embodiment of the present invention;
[0051] Figure 4 shows a schematic structural diagram of a terminal provided by an embodiment of the present invention. Detailed Embodiments
[0052] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art.
[0053] Aiming at the problem of low accuracy in identifying abnormal energy consumption structure data of existing enterprises. An embodiment of the present invention provides an abnormal identification method for energy consumption structure data, as Figure 1 shown, the method includes:
[0054] 101. Obtain the energy consumption data of various types of energy at different time nodes of the target enterprise within the historical period, and construct an energy consumption structure matrix based on the energy consumption data.
[0055] In the embodiment of the present invention, the target enterprise is an enterprise that involves the use of multiple types of energy and needs to analyze the structural distribution of the used energy. For example, chemical enterprises, steelmaking enterprises, petroleum refining enterprises, comprehensive energy service enterprises, etc. The energy types include but are not limited to coal, coke, petroleum, natural gas, and electricity. The historical period can be customized according to specific application scenario requirements, such as one month, six months, one year, etc., and the embodiment of the present invention does not make specific limitations. The energy consumption data is the energy consumption data corresponding to each type of energy at different time points, such as the daily energy consumption data.
[0056] In order to analyze the energy structure, it is necessary to count the energy proportion of different energies at different time nodes, and determine the proportion of the energy consumption of different energies in the total energy of the current time node at each time node. Taking a certain day as an example, the energy proportion of energy A is the proportion of the usage amount of energy A in the total usage amount of all energies on that day, and the energy consumption structure matrix includes the energy proportion of each energy type at each time node within the historical period. That is, the energy consumption structure matrix includes the energy proportion of each energy type at different time nodes.
[0057] 102. Use the trained energy consumption structure data reconstruction model to perform reconstruction processing on the energy consumption structure matrix to obtain an energy consumption structure reconstruction matrix.
[0058] In an embodiment of the present invention, an energy consumption structure data reconstruction model is pre-constructed, and a neural network model that has completed model training is included, which contains hidden layers, and the number of its hidden layers and the dimensional parameters of the model input and output are determined through adaptive adjustment. The energy consumption structure data reconstruction model learns the distribution characteristics of normal energy consumption structure data during the model training process, and can reconstruct the energy consumption structure matrix input into the model, so that the reconstructed matrix enables the energy consumption structure reconstruction matrix to meet the distribution of normal energy usage structure data. That is, the energy consumption structure reconstruction matrix is used to represent the energy consumption structure distribution under normal circumstances of the energy consumption structure. Through energy structure reconstruction, the distribution characteristics of the normal energy usage structure under the current energy usage type can be reflected. Since the energy usage types of different enterprises are not the same, by using the energy consumption data of the target enterprise itself for reconstruction, the energy consumption structure reconstruction matrix can be made to fit the energy usage characteristics of the current target enterprise to the greatest extent, providing more reasonable data support for subsequent anomaly identification.
[0059] 103. Determine the reconstruction error and the reconstruction error threshold based on the energy consumption structure matrix and the energy consumption structure reconstruction matrix.
[0060] In an embodiment of the present invention, since the energy consumption structure reconstruction matrix can reflect the distribution characteristics of the energy structure data of the target enterprise under normal circumstances, and the energy consumption structure matrix is the energy structure distribution data actually occurring in the target enterprise. Therefore, based on the difference between the two (reconstruction error), the deviation degree of the current actual energy structure data relative to the normal energy structure data can be characterized. And the deviation degree can be used as the direct basis for the abnormality of the energy consumption structure data. Of course, the existence of deviation does not necessarily mean abnormality, and it is also necessary to judge according to the magnitude of the deviation degree. Therefore, it is also necessary to determine a reconstruction error threshold to jointly use the reconstruction error threshold and the reconstruction error as the basis for anomaly judgment. Among them, the reconstruction error threshold can be customized according to experience or set according to the clustering or statistical results of the reconstruction error. The embodiment of the present invention does not make specific limitations.
[0061] 104. Generate an anomaly identification result of the energy consumption structure data of the target enterprise based on the reconstruction error and the reconstruction error threshold.
[0062] In the embodiments of the present invention, the abnormal recognition results of the target enterprise's energy consumption structure data include abnormal energy consumption structure points and abnormal energy consumption structure contexts. Among them, an abnormal energy consumption structure point refers to a situation where one or more time nodes in the energy consumption structure data are abnormal. For example, the reconstruction error at a certain time node is greater than the reconstruction error threshold, or the difference from the reconstruction error threshold is greater than the corresponding threshold. An abnormal energy consumption structure context refers to a situation where the change in the reconstruction error corresponding to adjacent time nodes in the energy consumption structure data is abnormal. For example, the difference in the reconstruction error corresponding to adjacent time nodes is greater than the corresponding threshold. Of course, the above two types of abnormalities can exist in parallel. By using the reconstruction error and the reconstruction error threshold, the abnormal energy consumption structure of the enterprise can be recognized from the single-point dimension of the deviation of the energy consumption structure data and the continuous change dimension of the energy consumption structure data, thereby improving the accuracy and comprehensiveness of energy consumption recognition.
[0063] It should be noted that by constructing an energy consumption structure matrix based on energy consumption data, reconstructing the energy consumption structure matrix based on a neural network model, and performing abnormal recognition of energy consumption structure data based on the reconstruction error, automatic abnormal recognition of energy consumption structure data is realized. It can timely discover abnormalities in the enterprise's energy use structure, help the enterprise optimize energy management, improve energy utilization efficiency, reduce energy waste, and improve economic benefits and environmental protection effects.
[0064] In one embodiment of the present invention, for further illustration and limitation, before the step of using the trained energy consumption structure data reconstruction model to reconstruct the energy consumption structure matrix to obtain an energy consumption structure reconstruction matrix, as Figure 2 shown, the method further includes:
[0065] 201. Obtain a normal energy consumption structure matrix sample, and construct an encoder and a decoder respectively.
[0066] 202. According to the normal energy consumption structure matrix sample input to the encoder, adaptively adjust the number of hidden layers and the input-output dimension parameters to obtain an initial energy consumption structure data reconstruction model.
[0067] 203. Construct an encoder loss function and a decoder loss function respectively, and determine the difference between the encoder loss function and the decoder loss function as a joint loss function.
[0068] 204. Use the joint loss function to train the initial energy consumption structure data reconstruction model to obtain a trained energy consumption structure data reconstruction model.
[0069] In the embodiments of the present invention, the energy consumption structure data reconstruction model includes an encoder and a decoder. The encoder is used to encode the input energy consumption structure matrix to obtain a latent variable, and the decoder is used to decode the latent variable to obtain the reconstructed energy consumption structure matrix. The normal energy consumption structure matrix sample is a screened and normal energy consumption structure matrix. The model is trained with this sample so that the model can learn the distribution characteristics of normal energy consumption structure data, and thus accurately reconstruct the energy consumption structure reconstruction matrix under normal conditions. According to the normal energy consumption structure matrix sample input to the encoder, adaptively adjusting the number of hidden layers and the input and output dimension parameters is to determine the initial structure of the model through adaptive adjustment in order to better adapt to the data characteristics.
[0070] By constructing loss functions for the encoder and the decoder respectively, the loss functions can better match the training objectives of the encoder and the decoder. The difference between the encoder loss function and the decoder loss function is determined as the joint loss function. When the difference between the results of the joint loss function in multiple rounds of training is less than the corresponding threshold, it indicates that the model parameters are relatively stable and the training can be determined to be completed.
[0071] In an embodiment of the present invention, for further illustration and limitation, the step of adaptively adjusting the number of hidden layers and the input and output dimension parameters according to the normal energy consumption structure matrix sample input to the encoder to obtain the initial energy consumption structure data reconstruction model includes:
[0072] According to the dimension of the normal energy consumption structure matrix sample and the preset latent variable distribution dimension, calculate the number of hidden layers of the model according to the hidden layer adaptive logarithmic function;
[0073] Calculate the dimensions of each hidden layer, the input dimension and the output dimension according to the number of hidden layers of the model to obtain the initial energy consumption structure data reconstruction model.
[0074] In the embodiments of the present invention, the dimension of the normal energy consumption structure matrix sample and the preset latent variable distribution dimension have a logarithmic function relationship with the number of hidden layers of the model. Let the dimension of the normal energy consumption structure matrix sample be n, and the preset latent variable distribution dimension be m. Then the calculation formula for the number of hidden layers K of the model, that is, the hidden layer adaptive logarithmic function, is expressed as: According to the obtained number of hidden layers of the model, the dimension of any hidden layer can be further calculated. The calculation formula is: h k = m × 2 k ; where h k represents the dimension of the kth hidden layer, k ∈ [1, K]. The input dimension of the encoder neural network is expressed as [n, h k , ……, h1, m], and the output dimension of the decoder neural network is expressed as [m, h1, ……, h k, n]. In one example, the dimension of the sample input model is n, and the dimension of the latent variable distribution is set to 5, then the number of hidden layers of the model Hidden layer h k has a dimension of 5 × 2 k , and the input dimension of the encoder neural network is [n, h k , ……, h1, 5], and the output dimension of the decoder neural network is [5, h1, ……, h k , n].
[0075] In an embodiment of the present invention, for further illustration and limitation, the construction process of the encoder loss function includes:
[0076] Using the encoder to encode the normal energy consumption structure matrix sample to obtain the posterior distribution of the latent variables of the normal energy consumption structure matrix sample;
[0077] Constructing an encoder loss function based on the KL divergence between the posterior distribution of the latent variables and the standard normal distribution;
[0078] The construction process of the decoder loss function includes:
[0079] Using the decoder to decode the latent variables of the normal energy consumption structure matrix sample to obtain the reconstruction result of the normal energy consumption structure matrix sample;
[0080] Constructing a decoder loss function based on the expectation of the log-likelihood of the reconstruction result on the posterior distribution of the latent variables, so as to balance the difference between the normal energy consumption structure matrix sample and the reconstruction result by taking the expectation of the posterior distribution of the latent variables.
[0081] In an embodiment of the present invention, after the encoder encodes the energy consumption structure matrix, the posterior distribution of the latent variables of the normal energy consumption structure matrix sample is obtained. This distribution describes the probability of possible values of the latent variables, specifically including the mean and standard deviation of the latent variable distribution. These two parameters are two key parameters describing the characteristics of the latent variable distribution; the mean represents the average level or central tendency of the latent variable, while the standard deviation measures the degree of dispersion or fluctuation of the latent variable distribution. These two parameters jointly determine the shape and position of the latent variable distribution. In Bayesian statistics, the posterior distribution refers to the update of the distribution of parameters or latent variables after given observed data. In this scenario, the normal energy consumption structure matrix sample is the given observed data. Since the posterior distribution of the latent variables of a normal energy consumption structure data should conform to the standard normal distribution, an encoder loss function can be constructed based on the KL divergence between the posterior distribution of the latent variables and the standard normal distribution. Among them, the KL divergence (Kullback-Leibler Divergence, KL) is a method for measuring the difference between two probability distributions. The smaller the KL divergence, the more similar the two distributions are. The encoder loss function is: KL(p θ (z|x q )||p(z)); where z is the latent variable, p θ (z|x q ) is the posterior distribution of the latent variable z under the condition of the normal energy consumption structure matrix sample x q , p(z) is the standard normal distribution of the latent variable z, where θ is the probability distribution parameter, and x q represents the q-th normal energy consumption structure matrix sample.
[0082] For the decoder, the purpose of the decoder is to make the reconstructed energy consumption structure matrix as close as possible to the original input energy consumption structure matrix. Therefore, a decoder loss function can be constructed based on the output of the decoder and the input of the encoder. The decoder loss function is expressed as: Among them, is the log-likelihood of the reconstruction result generated by the decoder; represents the expectation of p θ (z|x q ) for the latent variable z, is the log-likelihood parameter, which can be customized according to requirements.
[0083] Based on the above encoding loss function and decoding loss function, the joint loss function can be expressed as: Among them, Q is the number of samples.
[0084] In an embodiment of the present invention, for further illustration and limitation, the step of constructing the energy consumption structure matrix according to the energy consumption data includes:
[0085] Convert the energy consumption data according to the standard coal conversion coefficient corresponding to the energy type of the energy consumption data to obtain the standard coal consumption of each energy type at different time nodes;
[0086] For each energy type, calculate the proportion of the standard coal consumption under the energy type relative to the total standard coal consumption of the global energy type, and construct an energy consumption structure matrix based on the energy proportions of each energy type at different time nodes.
[0087] In the embodiment of the present invention, the energy consumption data is the usage data of any one energy corresponding to different time nodes of the target enterprise collected within the historical period. Taking the division unit of the time node as a natural day as an example, each time node corresponds to a natural day. Set D ij to represent the proportion of the consumption of the j-th energy of the enterprise on the i-th day in the total energy consumption on the i-th day:
[0088]
[0089] where i represents the date number, j represents the energy type number, and E ij represents the consumption of the j-th energy of the enterprise on the i-th day, and α j represents the standard coal conversion coefficient of the j-th energy.
[0090] In an embodiment of the present invention, for further illustration and limitation, the step of determining the reconstruction error and the reconstruction error threshold according to the energy consumption structure matrix and the energy consumption structure reconstruction matrix includes:
[0091] Calculate the differences between the energy proportions corresponding to the same energy type at the same time node in the energy consumption structure matrix and the energy consumption structure reconstruction matrix respectively, and sum the differences of different energy types at the same time node to obtain the reconstruction error at different time nodes;
[0092] Sort in descending order according to the reconstruction error values, and generate a reconstruction error curve with the sorting order of the reconstruction error as the horizontal axis and the value of the reconstruction error as the vertical axis;
[0093] Extract the inflection point corresponding to the largest slope of the broken line in the reconstruction error curve, and determine the reconstruction error corresponding to the inflection point as the reconstruction error threshold.
[0094] In the embodiment of the present invention, the reconstruction error is the sum of the errors of all energy types at different time nodes. The matrix dimensions of the energy consumption structure matrix and the energy consumption structure reconstruction matrix are the same. Subtract the elements (energy proportions) corresponding to the same time node and the same energy type in the two matrices, and sum the differences of all energy types at the same time node to obtain the reconstruction error. Set the energy consumption structure matrix to be represented as D ij , and the energy consumption structure reconstruction matrix to be represented as D ij, the calculation formula for the reconstruction error is as follows: where D ij represents the proportion of the consumption of the j-th type of energy by the enterprise on the i-th day to the total energy consumption on the i-th day, w is the number of energy types, and E i is the reconstruction error, and w is the total number of energy types on the i-th day. To extract the reconstruction error threshold based on the elbow method, the obtained reconstruction error values are sorted in descending order according to the values of the reconstruction error. Taking the sorting order of the reconstruction error as the horizontal axis and the value of the reconstruction error as the vertical axis, adjacent reconstruction errors are connected to form a broken line to construct a reconstruction error curve. Calculate the slopes of each broken line, and take the starting point corresponding to the broken line with the largest slope as the inflection point. The largest slope represents the fastest rate of decrease in the reconstruction error, that is, the point where the curve begins to become flat. Therefore, this is used as the reconstruction error threshold.
[0095] In an embodiment of the present invention, for further illustration and limitation, the step of generating an abnormal recognition result of the energy consumption structure data of the target enterprise based on the reconstruction error and the reconstruction error threshold includes:
[0096] Determine the time nodes corresponding to the reconstruction error greater than the reconstruction error threshold as abnormal energy consumption nodes;
[0097] If the number of abnormal energy consumption nodes within a preset time period is greater than the abnormal node threshold, determine that the abnormal recognition result of the energy consumption structure is an abnormal energy consumption structure set;
[0098] If the number of abnormal energy consumption nodes is less than or equal to the abnormal node threshold, determine that the abnormal recognition result of the energy consumption structure is an abnormal energy consumption structure point;
[0099] Sort the reconstruction errors in ascending order of time nodes. If the first reconstruction error at the previous order is greater than the second reconstruction error at the subsequent order among two adjacent reconstruction errors, and the absolute value of their difference is greater than a preset multiple of the first reconstruction error, determine that the abnormal recognition result of the energy consumption structure is an abnormal energy consumption structure context.
[0100] In an embodiment of the present invention, when there are time nodes with a reconstruction error greater than the reconstruction error threshold, it may include two situations: abnormal energy consumption structure set and abnormal energy consumption structure point. In the case of determining abnormal energy consumption nodes, it can be further determined whether it is an abnormal energy consumption structure set according to the comparison result between the number of abnormal energy consumption nodes within a preset time period and the abnormal node threshold. If it is not an abnormal energy consumption structure set, it is an abnormal energy consumption structure point. Among them, the abnormal node threshold is less than the number of time nodes included in the preset time period. Preferably, the preset time period is 7 days and the abnormal node threshold is 5. That is, sort the reconstruction error E i in ascending order of time, and slide and calculate the number of times S of point abnormalities in the daily energy consumption of the enterprise within every seven days i , which is expressed as a formula:
[0101]
[0102] When S i > 5, D i , ……, D i+7 When it is determined that the recognition result is an abnormal energy - using structure set, otherwise, it is an abnormal energy - using structure point.
[0103] In addition to the above two types of anomalies, the abnormal recognition result of the energy - using structure is an abnormal energy - using structure context. That is, arranged in ascending order of the time corresponding to the reconstruction error, if the first reconstruction error at the earlier pre - order position in the sequence is greater than the second reconstruction error at the later post - order position, and the absolute value of the difference between the two is greater than a preset multiple of the first reconstruction error, then it is determined that the abnormal recognition result of the energy - using structure is an abnormal energy - using structure context. Among them, the preset multiple is preferably 0.3, and of course, it can also be customized according to application requirements. Let the first reconstruction error be E i , and the second reconstruction error be E i+1 , when 0.7E i > E i+1 or 1.3E i < E i+1 In this case, the energy - using structure data corresponding to the time node of E i+1 is abnormal, and the recognition result is an abnormal energy - using structure context.
[0104] The present invention provides a method for abnormal recognition of energy - using structure data. In the embodiments of the present invention, by obtaining the energy - using data of various energy sources at different time nodes of a target enterprise during a historical period, an energy - using structure matrix is constructed based on the energy - using data, where the energy - using structure matrix includes the energy proportion of each energy type at different time nodes; the constructed energy - using structure matrix is reconstructed by using a trained energy - using structure data reconstruction model to obtain an energy - using structure reconstruction matrix, where the number of hidden layers and the input - output dimension parameters of the energy - using structure data reconstruction model are determined by adaptive adjustment, and the energy - using structure reconstruction matrix is used to represent the energy - using structure distribution under normal circumstances; the reconstruction error and the reconstruction error threshold are determined based on the energy - using structure matrix and the energy - using structure reconstruction matrix; an abnormal recognition result of the energy - using structure data of the target enterprise is generated based on the reconstruction error and the reconstruction error threshold, including abnormal energy - using structure points and abnormal energy - using structure contexts. By reconstructing the energy - using structure matrix based on a neural network model and using the reconstruction error as the basis for judging anomalies, the dependence on manual experience is greatly reduced, and the interference of subjective factors in manual analysis is reduced, thereby greatly improving the accuracy of abnormal recognition of energy - using structure data.
[0105] Furthermore, as for the above Figure 1For the implementation of the method described above, an embodiment of the present invention provides a recognition device for abnormal energy consumption structure data, as Figure 3 shown. The device includes:
[0106] An acquisition module 31, configured to acquire the energy consumption data of various types of energy at different time nodes of a target enterprise within a historical period, and construct an energy consumption structure matrix based on the energy consumption data, where the energy consumption structure matrix includes the energy proportion of each energy type at different time nodes;
[0107] A reconstruction module 32, configured to perform a reconstruction process on the energy consumption structure matrix by using a trained energy consumption structure data reconstruction model to obtain an energy consumption structure reconstruction matrix, where the number of hidden layers and the input-output dimension parameters of the energy consumption structure data reconstruction model are determined by adaptive adjustment, and the energy consumption structure reconstruction matrix is used to represent the energy consumption structure distribution under normal energy consumption structure conditions;
[0108] A determination module 33, configured to determine a reconstruction error and a reconstruction error threshold based on the energy consumption structure matrix and the energy consumption structure reconstruction matrix;
[0109] A generation module 34, configured to generate an abnormal recognition result of the energy consumption structure data of the target enterprise based on the reconstruction error and the reconstruction error threshold, including energy consumption structure point anomaly and energy consumption structure context anomaly.
[0110] Further, the device further includes:
[0111] A first construction module, configured to acquire a normal energy consumption structure matrix sample and construct an encoder and a decoder respectively;
[0112] An adjustment module, configured to adaptively adjust the number of hidden layers and the input-output dimension parameters according to the normal energy consumption structure matrix sample input to the encoder to obtain an initial energy consumption structure data reconstruction model;
[0113] A second construction module, configured to construct an encoder loss function and a decoder loss function respectively, and determine the difference between the encoder loss function and the decoder loss function as a joint loss function;
[0114] A training module, configured to train the initial energy consumption structure data reconstruction model by using the joint loss function to obtain a trained energy consumption structure data reconstruction model.
[0115] Further, the adjustment module includes:
[0116] A first calculation unit, configured to calculate the number of hidden layers of the model according to the adaptive logarithmic function of the hidden layer based on the dimension of the normal energy consumption structure matrix sample and the dimension of the preset latent variable distribution, wherein the dimension of the normal energy consumption structure matrix sample and the dimension of the preset latent variable distribution have a logarithmic function relationship with the number of hidden layers of the model;
[0117] A second calculation unit, configured to calculate the dimensions, input dimensions, and output dimensions of each hidden layer according to the number of hidden layers of the model, and obtain an initial energy consumption structure data reconstruction model.
[0118] Further, the second construction module includes:
[0119] An encoding unit, configured to encode the normal energy consumption structure matrix sample by using an encoder to obtain the posterior distribution of the latent variables of the normal energy consumption structure matrix sample;
[0120] A first construction unit, configured to construct an encoder loss function according to the KL divergence between the posterior distribution of the latent variables and the standard normal distribution;
[0121] A decoding unit, configured to decode the latent variables of the normal energy consumption structure matrix sample by using a decoder to obtain the reconstruction result of the normal energy consumption structure matrix sample;
[0122] A second construction unit, configured to construct a decoder loss function according to the expectation of the log-likelihood of the reconstruction result on the posterior distribution of the latent variables, so as to balance the difference between the normal energy consumption structure matrix sample and the reconstruction result by taking the expectation of the posterior distribution of the latent variables.
[0123] Further, the obtaining module 31 includes:
[0124] A conversion unit, configured to convert the energy consumption data according to the standard coal conversion coefficient corresponding to the energy type of the energy consumption data to obtain the standard coal consumption of each energy type at different time nodes;
[0125] A third calculation unit, configured to calculate, for each energy type, the proportion of the standard coal consumption of the energy type relative to the total standard coal consumption of the global energy type, and construct an energy consumption structure matrix according to the energy proportion of each energy type at different time nodes.
[0126] Further, the determining module 33 includes:
[0127] A fourth calculation unit, configured to calculate the difference between the energy proportions corresponding to the same energy type at the same time node in the energy consumption structure matrix and the energy consumption structure reconstruction matrix respectively, and sum the differences of different energy types at the same time node to obtain the reconstruction error at different time nodes.
[0128] A generating unit, configured to sort in descending order of the reconstruction error value, and generate a reconstruction error curve with the sorting order of the reconstruction error as the horizontal axis and the value of the reconstruction error as the vertical axis;
[0129] A first determining unit, configured to extract an inflection point with the largest broken-line slope in the reconstruction error curve, and determine the reconstruction error corresponding to the inflection point as the reconstruction error threshold.
[0130] Further, the generating module includes:
[0131] A second determining unit, configured to determine a time node corresponding to a reconstruction error greater than the reconstruction error threshold as an abnormal energy consumption node;
[0132] A third determining unit, configured to determine that the abnormal energy consumption structure recognition result is an abnormal energy consumption structure set if the number of abnormal energy consumption nodes within a preset time period is greater than the abnormal node threshold, where the abnormal node threshold is less than the number of time nodes included in the preset time period;
[0133] A fourth determining unit, configured to determine that the abnormal energy consumption structure recognition result is an abnormal energy consumption structure point if the number of abnormal energy consumption nodes is less than or equal to the abnormal node threshold;
[0134] A fifth determining unit, configured to sort the reconstruction errors in ascending order of time nodes. If a first reconstruction error at a previous order is greater than a second reconstruction error at a subsequent order among two adjacent-order reconstruction errors, and the absolute value of the difference between the two is greater than a preset multiple of the first reconstruction error, then determine that the abnormal energy consumption structure recognition result is an abnormal energy consumption structure context.
[0135] The present invention provides an abnormal recognition device for energy consumption structure data. In the embodiments of the present invention, by obtaining the energy consumption data of various types of energy at different time nodes of a target enterprise within a historical period, an energy consumption structure matrix is constructed based on the energy consumption data, wherein the energy consumption structure matrix includes the energy proportion of each energy type at different time nodes; the energy consumption structure matrix is reconstructed by using a trained energy consumption structure data reconstruction model to obtain an energy consumption structure reconstruction matrix, wherein the number of hidden layers and the input-output dimension parameters of the energy consumption structure data reconstruction model are determined by adaptive adjustment, and the energy consumption structure reconstruction matrix is used to represent the energy consumption structure distribution under normal circumstances; the reconstruction error and the reconstruction error threshold are determined based on the energy consumption structure matrix and the energy consumption structure reconstruction matrix; an abnormal recognition result of the energy consumption structure data of the target enterprise is generated based on the reconstruction error and the reconstruction error threshold, including energy consumption structure point abnormality and energy consumption structure context abnormality. By reconstructing the energy consumption structure matrix based on a neural network model and using the reconstruction error as the basis for judging abnormality, the dependence on manual experience is greatly reduced, and the interference of subjective factors in manual analysis is reduced, thereby greatly improving the accuracy of abnormal recognition of energy consumption structure data.
[0136] According to an embodiment of the present invention, a storage medium is provided. The storage medium stores at least one executable instruction, and the computer executable instruction can execute the abnormal recognition method of energy consumption structure data in any of the above method embodiments.
[0137] Figure 4 The structural schematic diagram of a terminal provided by an embodiment of the present invention is shown. The specific implementation of the terminal is not limited in the specific embodiments of the present invention.
[0138] As Figure 4 shown, the terminal may include: a processor 402, a communication interface 404, a memory 406, and a communication bus 408.
[0139] Among them: the processor 402, the communication interface 404, and the memory 406 communicate with each other through the communication bus 408.
[0140] The communication interface 404 is used for network communication with other devices such as clients or other servers.
[0141] The processor 402 is used to execute the program 410, and specifically can execute the relevant steps in the above embodiments of the abnormal recognition method of energy consumption structure data.
[0142] Specifically, the program 410 may include program codes, and the program codes include computer operation instructions.
[0143] The processor 402 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. One or more processors included in the terminal may be of the same type, such as one or more CPUs; or may be of different types, such as one or more CPUs and one or more ASICs.
[0144] The memory 406 is used to store the program 410. The memory 406 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk memory.
[0145] The program 410 is specifically configured to cause the processor 402 to perform the following operations:
[0146] Obtain the energy consumption data of various types of energy of the target enterprise at different time nodes within the historical period, and construct an energy consumption structure matrix based on the energy consumption data, where the energy consumption structure matrix includes the energy proportion of each energy type at different time nodes;
[0147] Use the trained energy consumption structure data reconstruction model to reconstruct the energy consumption structure matrix, and obtain an energy consumption structure reconstruction matrix, where the number of hidden layers and the input / output dimension parameters of the energy consumption structure data reconstruction model are determined by adaptive adjustment, and the energy consumption structure reconstruction matrix is used to represent the energy consumption structure distribution under normal circumstances;
[0148] Determine the reconstruction error and the reconstruction error threshold based on the energy consumption structure matrix and the energy consumption structure reconstruction matrix;
[0149] Generate an abnormal recognition result of the energy consumption structure data of the target enterprise based on the reconstruction error and the reconstruction error threshold, including energy consumption structure point abnormality and energy consumption structure context abnormality.
[0150] Obviously, those skilled in the art should understand that the above-mentioned modules or steps of the present invention can be implemented by a general-purpose computing device. They can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices. Optionally, they can be implemented by program codes executable by the computing device, so that they can be stored in the storage device and executed by the computing device. And in some cases, the steps shown or described can be executed in a different order than here, or they can be separately made into individual integrated circuit modules, or multiple modules or steps among them can be made into a single integrated circuit module to implement. Thus, the present invention is not limited to any specific combination of hardware and software.
[0151] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.
Claims
1. An abnormal recognition method for energy utilization structure data, characterized in that Including: Obtain the energy consumption data of various types of energy at different time nodes of the target enterprise within the historical period, and construct an energy consumption structure matrix based on the energy consumption data. Among them, the energy consumption structure matrix includes the energy proportion of each energy type at different time nodes; Use the trained energy consumption structure data reconstruction model to perform reconstruction processing on the energy consumption structure matrix to obtain an energy consumption structure reconstruction matrix. Among them, the number of hidden layers and the input-output dimension parameters of the energy consumption structure data reconstruction model are determined through adaptive adjustment, and the energy consumption structure reconstruction matrix is used to represent the energy consumption structure distribution under normal circumstances; Determine the reconstruction error and the reconstruction error threshold based on the energy consumption structure matrix and the energy consumption structure reconstruction matrix; Generate an abnormal recognition result of the energy consumption structure data of the target enterprise based on the reconstruction error and the reconstruction error threshold, including energy consumption structure point abnormality and energy consumption structure context abnormality.
2. The method according to claim 1, characterized in that, Before the step of using the trained energy consumption structure data reconstruction model to perform reconstruction processing on the energy consumption structure matrix to obtain an energy consumption structure reconstruction matrix, the method further includes: Obtain a normal energy consumption structure matrix sample, and respectively construct an encoder and a decoder; According to the normal energy consumption structure matrix sample input to the encoder, adaptively adjust the number of hidden layers and the input-output dimension parameters to obtain an initial energy consumption structure data reconstruction model; Respectively construct an encoder loss function and a decoder loss function, and determine the difference between the encoder loss function and the decoder loss function as a joint loss function; Use the joint loss function to train the initial energy consumption structure data reconstruction model to obtain a trained energy consumption structure data reconstruction model.
3. The method according to claim 2, wherein The step of adaptively adjusting the number of hidden layers and the input-output dimension parameters according to the normal energy consumption structure matrix sample input to the encoder to obtain an initial energy consumption structure data reconstruction model includes: According to the dimension of the normal energy consumption structure matrix sample and the preset latent variable distribution dimension, calculate the number of model hidden layers according to the hidden layer adaptive logarithmic function, where the dimension of the normal energy consumption structure matrix sample and the preset latent variable distribution dimension are in a logarithmic function relationship with the number of model hidden layers; Calculate the dimension, input dimension and output dimension of each hidden layer according to the number of model hidden layers to obtain an initial energy consumption structure data reconstruction model.
4. The method according to claim 2, characterized in that, The construction process of the encoder loss function includes: Use the encoder to encode the normal energy consumption structure matrix sample to obtain the posterior distribution of the latent variable of the normal energy consumption structure matrix sample; Construct an encoder loss function based on the KL divergence between the posterior distribution of the latent variable and the standard normal distribution; The construction process of the decoder loss function includes: Use the decoder to decode the latent variable of the normal energy consumption structure matrix sample to obtain the reconstruction result of the normal energy consumption structure matrix sample; Construct a decoder loss function based on the expectation of the log-likelihood of the reconstruction result on the posterior distribution of the latent variable, so as to balance the difference between the normal energy consumption structure matrix sample and the reconstruction result by taking the expectation of the posterior distribution of the latent variable.
5. The method according to claim 1, characterized in that, Constructing an energy consumption structure matrix based on the energy consumption data includes: Converting the energy consumption data according to the standard coal conversion coefficient corresponding to the energy type of the energy consumption data to obtain the standard coal consumption of each energy type at different time nodes; For each energy type, calculate the proportion of the standard coal consumption of the energy type relative to the total standard coal consumption of the global energy types, and construct an energy consumption structure matrix based on the energy proportions of each energy type at different time nodes.
6. The method according to claim 1, wherein Determining the reconstruction error and the reconstruction error threshold based on the energy consumption structure matrix and the energy consumption structure reconstruction matrix includes: Calculate the difference in the energy proportions corresponding to the same energy type at the same time node in the energy consumption structure matrix and the energy consumption structure reconstruction matrix respectively, and sum the differences of different energy types at the same time node to obtain the reconstruction error at different time nodes; Sort the reconstruction errors in descending order, and generate a reconstruction error curve with the sorting order of the reconstruction errors as the horizontal axis and the value of the reconstruction errors as the vertical axis; Extract the inflection point corresponding to the largest slope of the broken line in the reconstruction error curve, and determine the reconstruction error corresponding to the inflection point as the reconstruction error threshold.
7. The method according to claim 1, characterized in that Generating an abnormal recognition result of the energy consumption structure data of the target enterprise based on the reconstruction error and the reconstruction error threshold includes: Determine the time nodes corresponding to the reconstruction errors greater than the reconstruction error threshold as abnormal energy consumption nodes; If the number of the abnormal energy consumption nodes within a preset time period is greater than the abnormal node threshold, determine that the abnormal recognition result of the energy consumption structure is an abnormal energy consumption structure set, where the abnormal node threshold is less than the number of time nodes included in the preset time period; If the number of the abnormal energy consumption nodes is less than or equal to the abnormal node threshold, determine that the abnormal recognition result of the energy consumption structure is an abnormal energy consumption structure point; Sort the reconstruction errors in ascending order of time nodes. If the first reconstruction error at the previous order position is greater than the second reconstruction error at the subsequent order position among two adjacent order positions of the reconstruction errors, and the absolute value of the difference between the two is greater than a preset multiple of the first reconstruction error, determine that the abnormal recognition result of the energy consumption structure is an abnormal energy consumption structure context.
8. An identification device for abnormal energy-using structure data, characterized in that, Including: An acquisition module for acquiring the energy consumption data of various energy types of the target enterprise at different time nodes within a historical period, and constructing an energy consumption structure matrix based on the energy consumption data, where the energy consumption structure matrix includes the energy proportions of each energy type at different time nodes; A reconstruction module for reconstructing the energy consumption structure matrix by using a trained energy consumption structure data reconstruction model to obtain an energy consumption structure reconstruction matrix, where the number of hidden layers and the input-output dimension parameters of the energy consumption structure data reconstruction model are determined by adaptive adjustment, and the energy consumption structure reconstruction matrix is used to represent the energy consumption structure distribution under normal circumstances; A determination module for determining the reconstruction error and the reconstruction error threshold based on the energy consumption structure matrix and the energy consumption structure reconstruction matrix; A generation module for generating an abnormal recognition result of the energy consumption structure data of the target enterprise based on the reconstruction error and the reconstruction error threshold, including abnormal energy consumption structure points and abnormal energy consumption structure contexts.
9. A storage medium stores at least one executable instruction, and the executable instruction causes a processor to perform operations corresponding to the abnormal recognition method of the energy consumption structure data according to any one of claims 1-7.
10. A terminal, comprising: A processor, a memory, a communication interface, and a communication bus, and the processor, the memory, and the communication interface complete communication with each other through the communication bus; The memory is used for storing at least one executable instruction, and the executable instruction causes the processor to perform operations corresponding to the abnormal recognition method of the energy consumption structure data according to any one of claims 1-7.