CN116911414A - Power consumption prediction method, device, equipment and computer storage medium - Google Patents

Abstract

The embodiment of the invention relates to the technical field of computer data processing, and discloses a power consumption prediction method, which comprises the following steps: acquiring a power consumption prediction model corresponding to a user group where a user to be predicted is located; the electricity consumption prediction model is obtained by training a historical prediction model according to predicted electricity consumption data before a prediction period and real data of the prediction period; the loss function of the historical prediction model is determined according to predicted electricity consumption data before a prediction period, real data in the prediction period, electricity consumption behavior characteristics of a target user and weight information of the target user in a user group; the predicted electricity consumption data before the prediction period is the predicted electricity consumption data of the target user in the prediction period, which is output by the history prediction model; the real data of the prediction period is the real electricity consumption data of the target user in the prediction period; and inputting the historical electricity consumption data of the user to be predicted into the electricity consumption prediction model to obtain the predicted electricity consumption data of the user to be predicted. By the mode, the power consumption prediction accuracy is improved.

Description

Power consumption prediction method, device, equipment and computer storage medium

Technical Field

The embodiment of the invention relates to the technical field of computer data processing, in particular to a power consumption prediction method, a device, equipment and a computer storage medium.

Background

As the population grows and business activities proceed, the demand for electricity continues to increase. Therefore, prediction of the amount of electricity consumption becomes also a key in the electricity supply process.

The inventors of the present application found in the course of carrying out the embodiment of the present method that: the existing electricity consumption prediction only normalizes the collected steady-state statistical data and transient data, predicts according to the characteristic information related to the electricity consumption, and has the problem of low prediction accuracy.

Disclosure of Invention

In view of the above problems, an embodiment of the present application provides a power consumption prediction method, which is used to solve the problem in the prior art that the accuracy of power consumption prediction is low.

According to an aspect of an embodiment of the present application, there is provided a power consumption prediction method, including:

acquiring a power consumption prediction model corresponding to a user group where a user to be predicted is located; the electricity consumption prediction model is obtained by training a historical prediction model according to predicted electricity consumption data before a prediction period and real data of the prediction period; the loss function of the history prediction model is determined according to the predicted electricity consumption data before the prediction period, the real data in the prediction period, the electricity consumption behavior characteristics of the target user and the weight information of the target user in the user group; the predicted electricity consumption data before the prediction period is the predicted electricity consumption data of the target user in the prediction period, which is output by the history prediction model; the prediction period real data are real electricity utilization data of the target user in the prediction period;

And inputting the historical electricity consumption data of the user to be predicted into the electricity consumption prediction model to obtain the predicted electricity consumption data of the user to be predicted.

In an alternative, the method further comprises:

determining cross entropy between the predicted electricity consumption data before the prediction period and the real data in the prediction period;

determining cross entropy correction parameters corresponding to the user group according to the electricity behavior characteristics and the weight information;

and determining the loss function according to the cross entropy, the cross entropy correction parameter and a preset cross entropy threshold.

In an alternative manner, the cross entropy correction parameter comprises a first correction parameter; the method further comprises the steps of:

determining the acquired data extreme value and the mean value information of the target user according to the electricity behavior characteristics;

and determining the first correction parameter according to the acquired data extreme value and the mean value information.

In an alternative manner, the cross entropy correction parameter comprises a second correction parameter; the target users comprise users with electricity consumption and data acquisition capacity meeting preset conditions in the user group; the method further comprises the steps of:

determining the weight information according to the proportion of the target user to the total number of users in the user group;

And determining the second correction parameter according to the weight information.

In an alternative, the method further comprises:

correcting the cross entropy according to the cross entropy correction parameter to obtain corrected cross entropy;

and determining the loss function according to the comparison result of the corrected cross entropy and the cross entropy threshold.

In an alternative manner, the user group is one of a plurality of alternative groups; the plurality of selectable groups are obtained by dividing all users in the target area according to the user characteristic images; the history prediction model comprises an updated global model; the method further comprises the steps of:

the initial prediction model is issued to the edge side server corresponding to each optional group;

for each optional group, training the initial prediction model through the edge side server according to the predicted electricity consumption data before the prediction period and the predicted real data in the prediction period corresponding to the optional group to obtain an updated group model corresponding to the optional group;

the updated group models returned by all the edge side servers are aggregated according to the group characteristic information of the selectable groups, so that updated global models are obtained;

And sending the updated global model to each edge server so that the edge server carries out the next training on the updated global model.

In an alternative, the method further comprises:

determining group data reference weights corresponding to the optional groups and similarity among the optional groups according to the group characteristic information;

and aggregating the updated group model according to the similarity and the group data reference weight to obtain the updated global model.

According to another aspect of the embodiment of the present invention, there is provided a power consumption prediction apparatus including:

the acquisition module is used for acquiring a power consumption prediction model corresponding to a user group where a user to be predicted is located; the electricity consumption prediction model is obtained by training a historical prediction model according to predicted electricity consumption data before a prediction period and real data of the prediction period; the loss function of the history prediction model is determined according to the predicted electricity consumption data before the prediction period, the real data in the prediction period, the electricity consumption behavior characteristics of the target user and the weight information of the target user relative to the user group; the predicted electricity consumption data before the prediction period is the predicted electricity consumption data of the target user in the prediction period, which is output by the history prediction model; the prediction period real data are real electricity utilization data of the target user in the prediction period;

And the prediction module is used for inputting the historical electricity consumption data of the user to be predicted into the electricity consumption prediction model to obtain the predicted electricity consumption data of the user to be predicted.

According to another aspect of the embodiment of the present invention, there is provided a power consumption prediction apparatus including: the device comprises a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface complete communication with each other through the communication bus;

the memory is configured to store at least one executable instruction that causes the processor to perform the operations of any of the power usage prediction method embodiments described above.

According to yet another aspect of embodiments of the present invention, there is provided a computer-readable storage medium having stored therein at least one executable instruction for performing the operations of any of the power consumption prediction method embodiments using a power consumption prediction device.

The embodiment of the invention obtains the power consumption prediction model corresponding to the user group where the user to be predicted is located; the electricity consumption prediction model is obtained by training a historical prediction model according to predicted electricity consumption data before a prediction period and real data of the prediction period; the loss function of the history prediction model is determined according to the predicted electricity consumption data before the prediction period, the real data in the prediction period, the electricity consumption behavior characteristics of the target user and the weight information of the target user in the user group; the target users can be users meeting preset electricity consumption and data acquisition conditions in the user group and are used for representing electricity consumption characteristics of the whole user group, and the history prediction model is trained through electricity consumption data acquired from the selected target users, so that model training effect is ensured, and model training efficiency is improved. Further, considering the influence of the selected target user on model training, the difference between the predicted electricity consumption data before the prediction period and the predicted real data can be corrected according to the electricity consumption behavior characteristics of the target user and the weight of the target user occupying the user group, and a loss function is calculated according to the corrected difference, so that the electricity consumption prediction model is optimized. The electricity consumption behavior characteristics of the target user and the predicted electricity consumption data of the target user in the prediction period, which occupy the user group, are output by the history prediction model; the prediction period real data are real electricity utilization data of the target user in the prediction period; the historical electricity consumption data of the user to be predicted is input into the electricity consumption prediction model to obtain the predicted electricity consumption data of the user to be predicted, the predicted electricity consumption data is different from the whole steady state data selected from the power grid in the prior art to conduct electricity consumption prediction, the efficiency is low, the influence of the electricity consumption behavior of the user on a model prediction result is ignored, the power grid user is divided into a plurality of user groups according to user attributes, each user group corresponds to a specific electricity consumption prediction model, and the result of model training is corrected according to the electricity consumption behavior characteristics of a target user selected in the user group and the relation between the power consumption behavior characteristics and the whole group, so that the efficiency and the accuracy of electricity consumption prediction are improved.

The foregoing description is only an overview of the technical solutions of the embodiments of the present invention, and may be implemented according to the content of the specification, so that the technical means of the embodiments of the present invention can be more clearly understood, and the following specific embodiments of the present invention are given for clarity and understanding.

Drawings

The drawings are only for purposes of illustrating embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to designate like parts throughout the figures. In the drawings:

fig. 1 is a schematic flow chart of a power consumption prediction method according to an embodiment of the present invention;

fig. 2 is a schematic structural diagram of a power consumption prediction device according to an embodiment of the present invention;

fig. 3 is a schematic structural diagram of a power consumption prediction apparatus according to an embodiment of the present invention.

Detailed Description

Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention may be embodied in various forms and should not be limited to the embodiments set forth herein.

Prior art and its problems are further described before proceeding with the description of the embodiments of the present invention:

The power consumption prediction is influenced by various factors, the prior art scheme only normalizes the collected steady-state statistical data and transient data and then carries out prediction work, and the influence of the maximum value and the minimum value of the power consumption on the prediction is ignored; in addition, in the prior art, different subareas in the whole prediction area are not divided according to consumption types, and meanwhile, the proportion of the monitoring equipment quantity of each subarea to the total equipment quantity of the area is not taken into consideration; finally, the electricity consumption prediction is closely related to the behavior of the user, and the prior art scheme ensures the privacy problem of the user by utilizing the federal learning technology, but lacks accurate and effective parameter indexes for measuring the relation between the predicted electricity consumption and the actual electricity consumption, so that the model effect cannot be well evaluated.

Brief description of related nouns:

the edge computing technology is a distributed open platform integrating network, computing, storage and application core capabilities at the network edge side close to an object or data source, provides edge intelligent service nearby, and meets key requirements of the power industry in aspects of agility connection, real-time service, data optimization, application intelligence and the like.

Federal learning techniques support model learning, training, and predictive work in a distributed manner on multiple devices or servers. Meanwhile, in the process, all model training data can be stored on the local equipment without following the traditional machine learning mode, and the data is stored on the centralized entity, so that the information safety is fully ensured.

Fig. 1 shows a flowchart of a power consumption prediction method provided by an embodiment of the present invention, where the method is executed by a computer processing device. The computer processing device may include a cell phone, a notebook computer, etc. As shown in fig. 1, the method comprises the steps of:

step 10: acquiring a power consumption prediction model corresponding to a user group where a user to be predicted is located; the electricity consumption prediction model is obtained by training a historical prediction model according to predicted electricity consumption data before a prediction period and real data of the prediction period; the loss function of the history prediction model is determined according to the predicted electricity consumption data before the prediction period, the real data in the prediction period, the electricity consumption behavior characteristics of the target user and the weight information of the target user in the user group; the predicted electricity consumption data before the prediction period is the predicted electricity consumption data of the target user in the prediction period, which is output by the history prediction model; and the prediction period real data is real electricity utilization data of the target user in the prediction period.

The user groups are obtained by dividing the power grid users according to the user feature portraits, for example, the user groups can be created by combining a group of users with similar user feature attributes, such as users with geographic proximity. The target user is a plurality of edge side devices which are selected from the user group where the user to be predicted is located and used for representing the user group, and the power consumption prediction model corresponding to the user group is trained by collecting power consumption data of the target user. Therefore, the target user may be an electricity terminal, such as a smart meter, capable of satisfying a sufficient amount of electricity data collection and expanding the scale of training data.

The prediction period may be a specific time interval for which the model training process is aimed, and the historical prediction model predicts the electricity consumption data of the target user in the prediction period before the prediction period. The electricity consumption data may include information such as electricity consumption, electricity consumption frequency, and peak value.

The historical prediction model can be a power consumption prediction model obtained by training in the previous training period, or an initial prediction model. Optionally, the historical prediction model may also be obtained by performing federal learning on a plurality of user groups included in the power grid, where each user group trains the model by using locally acquired electricity data of a specific client in an edge calculation manner, and uploads the training result to a cloud server, the cloud server gathers and aggregates the training results of all user groups to obtain an updated global model, and the updated global model is used as an initial prediction model of a new training process and is issued to an edge server corresponding to each user group, so that each edge server trains the initial prediction model according to the locally newly acquired data to obtain an electricity consumption prediction model corresponding to the group.

The optimization goal of the electricity consumption prediction model is that the output predicted electricity consumption data before the prediction period is infinitely close to the real data in the prediction period. Thus, the loss function of the historical prediction model may be characterized in terms of the gap between the predicted electricity data before the prediction period and the real data during the prediction period, and further, the gap may be characterized by cross entropy. Further, in order to improve the training efficiency of the power consumption prediction model corresponding to the user group, a part of representative target users are selected from the user group, and training of the model is performed according to the power consumption data collected by the target users, specifically, the target users can be users with power consumption reaching a certain level and data collection capacity meeting requirements, so that sufficient and continuous data can be ensured to be used for model training.

The selection of the target user is considered to determine the representativeness of the electricity consumption behavior of the user level to the whole user group, so that the electricity consumption behavior characteristics of the target user and the weight information of the target user occupying the user group need to be further considered when the training effect of the model is evaluated, and the difference between the predicted electricity consumption data before the prediction period and the predicted real data is corrected according to the electricity consumption behavior characteristics and the weight information, so that the actual electricity consumption situation of the whole user group is more met. The weight information is used for representing the representativeness of the target user to the user group where the target user is located, and can be determined according to the quantity proportion of the target user to the user group, the user importance level of the target user and the like.

Thus, in one embodiment of the present invention, prior to step 10, further comprising:

step 101: and determining the cross entropy between the prediction electricity consumption data before the prediction period and the prediction period real data.

Specifically, the calculation mode of the cross entropy can refer to the existing calculation mode, in an alternative mode, the Monte Carlo method adopted in the power system evaluation is combined, the predicted electricity consumption number before the prediction period and the real data in the prediction period can be sampled respectively, the corresponding probability density functions are obtained respectively, and the cross entropy is further calculated according to the probability density functions.

Step 102: and determining the cross entropy correction parameters corresponding to the user group according to the electricity behavior characteristics and the weight information.

Specifically, the cross entropy correction parameter includes a first correction parameter and a second correction parameter. In consideration of the fact that abnormal conditions in data, such as more sporadic extreme power consumption data deviating from the average power consumption level, affect the convergence efficiency of model training when training the model according to collected historical power consumption data of the user, extreme data in the target user power consumption data need to be screened according to user behavior characteristics, and a first correction parameter is determined according to the influence degree of the extreme power consumption data on cross entropy. Optionally, determining the influence weight of the acquired data of the target user on the cross entropy according to the weight information, and obtaining the second correction parameter. The weight information may be determined according to a ratio of the number of target users to the total number of users in the user group in which the target users are located.

Thus, in yet another embodiment of the present invention, step 102 further comprises:

step 1021: and determining the acquired data extreme value and the mean value information of the target user according to the electricity behavior characteristics.

In the embodiment of the invention, the average value of the collected data of the target user is determined to be according to the electricity behavior characteristics of the target user, which is different from the fact that in the prior art, when the electricity consumption is predicted, only more stable historical electricity consumption data is considered, and the influence of the maximum value and the minimum value of the electricity consumption data of the selected data collection object on the model training efficiency and the prediction accuracy is ignoredThe maximum value and the minimum value are p respectively _max And p _min 。

Step 1022: and determining the first correction parameter according to the acquired data extreme value and the mean value information.

Specifically, the first correction parameter R _A1 The calculation can be performed as follows:

in yet another embodiment of the present invention, in addition to extreme conditions in the power usage behavior characteristics of the target user itself affecting the reliability of the target user's characterization of the user group in which it is located, the user count ratio of the target user to the user group, etc., may also affect the data weight of the target user.

Thus, optionally, the cross-entropy correction parameter comprises a second correction parameter; the target users comprise users with electricity consumption and data acquisition capacity meeting preset conditions in the user group; as described above, in order to improve the efficiency of the power consumption prediction model training for the user group on the edge side, a part of users having a representative and abundant data amount may be selected as the acquisition target of the model training sample data.

Step 102 further comprises:

step 1023: and determining the weight information according to the proportion of the target user to the total number of users in the user group.

Specifically, taking the user group a as an example, considering the proportion of the equipment amount corresponding to the target user of the group a to the overall equipment amount of the area where the group a is located, and regarding the proportion as a representative representation of the overall electricity consumption data of the target user to the overall electricity consumption data of the user group a, so as to appropriately correct the cross entropy according to the calculation result of the proportion. It will be readily appreciated that in alternative embodiments of the present invention, the weight information of the target user may also be determined based on the user profile characteristics of the target user itself, such as the importance level of the user, where the importance level may be determined based on the power usage scale of the user.

Step 1024: and determining the second correction parameter according to the weight information.

Specifically, the ratio may be directly determined as the second correction parameter. I.e. defining the group a monitoring device quantity as N _A The overall monitoring equipment quantity of the area where the group A is located is N, and S is defined _A For the second correction parameter, the formula is as follows:

step 103: and determining the loss function according to the cross entropy, the cross entropy correction parameter and a preset cross entropy threshold.

Specifically, the cross entropy may be corrected according to the cross entropy correction parameter so that the corrected cross entropy can reflect the influence of the selection of the target user as the acquisition object of the sample data of the training model on the model training effect. And comparing the corrected cross entropy with a cross entropy threshold value, wherein the cross entropy threshold value is used for representing the cross entropy value of which the prediction accuracy reaches the model training completion requirement, so that the difference value between the corrected cross entropy and the cross entropy threshold value can be used for determining a loss function.

In yet another embodiment of the present invention, step 103 further comprises:

step 1031: and correcting the cross entropy according to the cross entropy correction parameter to obtain corrected cross entropy.

Specifically, in combination with the calculation manner of the cross entropy parameter, the product of the cross entropy correction parameter and the cross entropy may be determined as the corrected cross entropy, where the cross entropy parameter may be the aforementioned first correction parameter, second correction parameter, or the product of the first correction parameter and the second correction parameter.

Step 1032: and determining the loss function according to the comparison result of the corrected cross entropy and the cross entropy threshold.

Specifically, the difference between the corrected cross entropy and the cross entropy threshold may be determined as a function value of the loss function, wherein the smaller the corrected cross entropy is, the smaller the loss function is, and the closer the loss function is to the optimization target of the electricity consumption prediction model.

In yet another embodiment of the present invention, the user group is one of a plurality of selectable groups; the plurality of selectable groups are obtained by dividing all users in the target area according to the user characteristic images; the history prediction model comprises an updated global model; step 10 further comprises:

step 110: and issuing the initial prediction model to the edge side server corresponding to each optional group.

Specifically, the initial prediction model may be randomly generated by the cloud server or obtained by training a preset global model in advance according to publicly available historical electricity consumption data, and is used as a basis of overall prediction work. In particular, the historical power usage data may be a power value o= { o1, o2, on, the data volume must be guaranteed to be more than 20 numerical values, and the monitoring frequency can realize sufficient technical guarantee for the accuracy of power consumption monitoring. The cloud server can be used as a center for data summarization and data storage, and the cloud server stores historical prediction electricity consumption data of all user groups and model parameters of a historical electricity consumption prediction model.

Step 111: and training the initial prediction model according to the predicted electricity consumption data before the prediction period and the predicted real data corresponding to the optional groups through the edge side server aiming at each optional group to obtain updated group models corresponding to the optional groups.

Specifically, the process of training the initial prediction model by the edge server may refer to the foregoing step 10, that is, the training target of the initial prediction model is that the predicted power consumption data before the prediction period approaches the predicted period real data infinitely, the loss function of the initial prediction model is determined according to the predicted power consumption data before the prediction period and the difference between the cross entropy of the predicted period real data and the preset cross entropy threshold, where the cross entropy is corrected according to the power consumption behavior characteristics of the training sample acquisition object of the optional group and the weight of the optional group.

Step 112: and aggregating all the updated group models returned by the edge side servers according to the group characteristic information of the optional group to obtain the updated global model.

Specifically, model parameters of the updated group model returned by all edge side servers can be collected through the cloud server, optional groups are clustered according to group feature information, and model parameters corresponding to the same group are aggregated, so that the efficiency of model training is further improved through sharing and collaborative training of model parameters among similar user groups.

Thus, in yet another embodiment of the present invention, step 112 further comprises:

step 1121: and determining group data reference weights corresponding to the optional groups and the similarity among the optional groups according to the group characteristic information.

Specifically, the group characteristic information includes user characteristic information in the optional group, such as region information, type of electric equipment, and power consumption scale. And determining the demand level of electricity utilization guarantee and the user scale of the users of the optional group according to the group characteristic information, and determining the group data reference weight according to the group characteristic information. It is easy to understand that the larger the electricity consumption scale of a group, the higher the demand for electricity usage guarantee, the larger the group data reference weight of the group. For example, for a group corresponding to a large residential community or for a relatively important electricity utilization unit such as a data center, a core base station and the like, the group data reference weight is relatively large, so that the accuracy of electricity consumption prediction of the group is ensured, and an effective basis is provided for subsequent electricity utilization planning and guarantee. And carrying out cluster analysis on all the optional groups according to the group characteristic information to obtain the similarity among the optional groups.

Step 1122: and aggregating the updated group model according to the similarity and the group data reference weight to obtain the updated global model.

Specifically, the selectable groups with similarity higher than a preset similarity threshold may be classified into a group class, and weighted voting is performed according to model parameters of the updated group model corresponding to the selectable groups of the group class according to the group data reference weight, so as to obtain the updated global model.

Step 113: and sending the updated global model to each edge server so that the edge server carries out the next training on the updated global model.

Specifically, the updated global model is issued to each edge server through the cloud server. And the edge server performs new round of training on the updated global model according to the steps, and uploads the model obtained by the current round to the cloud server again after the training is completed to aggregate training results and prediction results, so that the model training is performed by locally utilizing an edge computing technology through a plurality of optional groups, the training results of all the optional groups are aggregated by utilizing a federal learning technology, and the efficiency and the effect of the power consumption prediction model training are improved.

Step 20: and inputting the historical electricity consumption data of the user to be predicted into the electricity consumption prediction model to obtain the predicted electricity consumption data of the user to be predicted.

Specifically, considering that the users in one user group have similarity in electricity consumption behavior characteristics, the electricity consumption prediction model can predict electricity consumption of the users in one user group, and predicted electricity consumption data comprises predicted electricity consumption of the users to be predicted.

In still another embodiment of the present invention, the process of performing the electricity consumption prediction may further include:

in combination with predicting short-term severe weather fluctuations and changing season effects over a period of time, more stable power usage monitoring data is combined for a group of users with similar attributes (e.g., geographic proximity) and created as a user group. For the federal learning process for a user group, we develop through basic data provided by edge devices (e.g., smart meters, etc.) deployed at the end-user side of the grid. By the method, a federal learning test and training enabling structure using edge equipment in the intelligent power grid can be established, and meanwhile, the accuracy of electricity consumption prediction is evaluated by a method for calculating cross entropy parameters.

Step 1: on the edge side server of each user group, the initialized global model is trained in advance according to publicly available historical electricity consumption data, or an initialized global model is randomly generated to serve as the basis of the whole prediction work. The definition of the global model is the power value o= { o1 at a certain moment, o2, on, the data size must be guaranteed to be above 20 values, the monitoring frequency is the accuracy of power consumption monitoring, and the sufficient technical guarantee is realized.

Step 2: in each designated training round, the edge side server of each user group sends the global model to the client end with abundant electricity consumption data acquisition quantity and can meet the two requirements of expanding the training data scale, and the aim of the step is to enable people to have enough data to reflect the electricity consumption condition of the user group.

Step 3: the definition of the electricity collection data is that the power value p= { p1 collected at a certain moment, p2, p.pn, based on the characteristic of easy uncertainty of electricity consumption, the data volume is required to be ensured to be more than 20 values, and the monitoring frequency is the accuracy of power consumption monitoring, so that sufficient technical guarantee is realized.

Step 4: assuming that user group A has a client, based on the power consumption data it collects, a batch gradient descent algorithm (BGD) is used to calculate an average gradient G _A1 Learning rate k _A1 Obtaining an updated model L _A1 。

Step 5: the client updates the model L _A1 And then, the model is sent to the edge side server of the user group A, and a brand new model applicable to the user group is obtained.

Step 6: and carrying out cross entropy numerical calculation on the brand new model and the finally counted real electricity consumption data. The cross entropy is used for judging the mutual support degree between two information sources and determining the weight of the information sources according to the mutual support degree. Furthermore, a probability density function is established to form a cross entropy objective function, weight parameters are determined, and then the model is iterated according to the weight parameters, so that the model is continuously approximate to a real electricity consumption value.

Step 7: to the model L _A1 Is sampled, which is expected to be defined as E _A1 (F)

Density function parameters.

Step 8: defined herein as g _A1 (x; u) it and F _A1 (x) Having the same distribution but being a probability density function of different parameters, further we define f _A1 (x; u) and g _A1 The ratio of (x; u) is a likelihood ratio function W _A1 (x; u) is:

step 9: substituting the formula in step 8 into step 7 and defining a new expected value J _A1 The formula can be rewritten as:

step 10: when 1 prediction can reach J _A1 True value of (1), i.e. J _A1 Where the variance of (g) is 0, g _A1 (x; u) satisfies the following condition:

in the above-mentioned description of the invention,is defined as g _A1 (x; u), wherein v is a parameter corresponding to the optimal sampling probability function.

Step 11: due to J _A1 As a result of the unknown quantity,and is not directly available, so we use a method of calculating cross entropy to solve. The smaller the value of the cross entropy, the more accurate the prediction accuracy of the electricity consumption of the model LA 1.

Step 12: considering that only more stable data is considered in the prior power consumption and the influence of the maximum value and the minimum value on prediction is ignored, we set the average value of a group of collected power values asAnd the maximum and minimum of them are p respectively _max And p _min . Thus, for this predictive model, R is defined _A1 Correcting parameters for the mean of cross entropy (i.e. the firstA correction parameter) as follows:

step 13: except the influence of the maximum value and the minimum value of the power, the proportion of the group A monitoring equipment quantity to the whole monitoring equipment quantity of the area where the group A is positioned is considered again, the cross entropy is properly corrected according to the calculation result of the proportion, and the group A monitoring equipment quantity is defined as N _A The overall monitoring equipment quantity of the area where the group A is located is N, and S is defined _A For the monitor device proportion correction parameter (i.e. the aforementioned second correction parameter) of the cross entropy, the formula is as follows:

step 14: after obtaining the above two cross entropy correction parameters, we get g based on the basic definition of cross entropy and considering the correction parameters again _A1 (x; u) andthe cross entropy of (a) is as follows:

step 15:this term is constant. In the process of the electricity consumption prediction, we will depend on the upper limit epsilon of the cross entropy and S _A ，R _A1 And +.>The values of these three terms calculate the cross entropy +.>And comparing with epsilon to judge whether the accuracy of the predicted value meets the requirement. If the finally obtained cross entropy result is smaller than epsilon, the accuracy of the predicted value meets the requirement, the federal learning process can be ended, and the predicted electricity consumption of the group A and the corresponding predicted model parameters are reported to the corresponding edge side servers.

Step 16: if the finally obtained cross entropy result is larger than or equal to epsilon, the accuracy of the predicted value is not satisfactory. At this point, further iterations are required, i.e. the whole process from step 4 to step 15 is repeated until the final cross entropy result is less than ε. And after the accuracy meets the requirement, reporting the predicted electricity consumption of the group A to the corresponding edge side server.

Step 17: in the prediction area where the group A is located, the prediction process of the group A is used as a reference, the same electricity consumption prediction process is respectively carried out on the group B, the group C, the group D and the like, and the electricity consumption prediction results obtained by the group B, the group C, the group D and the like are respectively reported to the corresponding edge side servers, so that preliminary summarization of the electricity consumption of each group in the area is realized.

Step 18: and after the primary summary is completed, all the edge side servers send the power consumption prediction data to the cloud server. And the cloud server accumulates sufficient quantity of electricity consumption data of each group in a period of time, and performs updating work of initializing a global model of each group in the area. After the updating work is completed, the cloud server distributes the initialized global model of each group to each edge measuring server again, then calculates average gradient and learning rate by using a batch gradient descent algorithm (BGD) according to the electricity consumption data collected by the corresponding groups to obtain an updated model, and judges the closeness degree of the electricity consumption predicted value and the real electricity consumption data by using an edge calculation and bang learning technology and improving a cross entropy algorithm, namely, repeatedly carries out a new round of prediction work. After each group meets the requirements, the updated data is sent to the cloud server again, and the global model is updated, so that a closed loop flow is formed.

The embodiment of the invention obtains the power consumption prediction model corresponding to the user group where the user to be predicted is located; the electricity consumption prediction model is obtained by training a historical prediction model according to predicted electricity consumption data before a prediction period and real data of the prediction period; the loss function of the history prediction model is determined according to the predicted electricity consumption data before the prediction period, the real data in the prediction period, the electricity consumption behavior characteristics of the target user and the weight information of the target user in the user group; the target users can be users meeting preset electricity consumption and data acquisition conditions in the user group and are used for representing electricity consumption characteristics of the whole user group, and the history prediction model is trained through electricity consumption data acquired from the selected target users, so that model training effect is ensured, and model training efficiency is improved. Further, considering the influence of the selected target user on model training, the difference between the predicted electricity consumption data before the prediction period and the predicted real data can be corrected according to the electricity consumption behavior characteristics of the target user and the weight of the target user occupying the user group, and a loss function is calculated according to the corrected difference, so that the electricity consumption prediction model is optimized. The electricity consumption behavior characteristics of the target user and the predicted electricity consumption data of the target user in the prediction period, which occupy the user group, are output by the history prediction model; the prediction period real data are real electricity utilization data of the target user in the prediction period; the historical electricity consumption data of the user to be predicted is input into the electricity consumption prediction model to obtain the predicted electricity consumption data of the user to be predicted, the predicted electricity consumption data is different from the whole steady state data selected from the power grid in the prior art to conduct electricity consumption prediction, the efficiency is low, the influence of the electricity consumption behavior of the user on a model prediction result is ignored, the power grid user is divided into a plurality of user groups according to user attributes, each user group corresponds to a specific electricity consumption prediction model, and the result of model training is corrected according to the electricity consumption behavior characteristics of a target user selected in the user group and the relation between the power consumption behavior characteristics and the whole group, so that the efficiency and the accuracy of electricity consumption prediction are improved.

Fig. 2 is a schematic structural diagram of a power consumption prediction apparatus according to an embodiment of the present invention. As shown in fig. 2, the apparatus 30 includes: an acquisition module 301 and a prediction module 302.

The acquiring module 301 is configured to acquire a power consumption prediction model corresponding to a user group where a user to be predicted is located; the electricity consumption prediction model is obtained by training a historical prediction model according to predicted electricity consumption data before a prediction period and real data of the prediction period; the loss function of the history prediction model is determined according to the predicted electricity consumption data before the prediction period, the real data in the prediction period, the electricity consumption behavior characteristics of the target user and the weight information of the target user relative to the user group; the predicted electricity consumption data before the prediction period is the predicted electricity consumption data of the target user in the prediction period, which is output by the history prediction model; the prediction period real data are real electricity utilization data of the target user in the prediction period;

and the prediction module 302 is configured to input the historical electricity consumption data of the user to be predicted into the electricity consumption prediction model, so as to obtain the predicted electricity consumption data of the user to be predicted.

The execution process of the electricity consumption prediction device provided by the embodiment of the invention is approximately the same as that of the foregoing method embodiment, and will not be repeated.

The electricity consumption prediction device provided by the embodiment of the invention obtains the electricity consumption prediction model corresponding to the user group where the user to be predicted is located; the electricity consumption prediction model is obtained by training a historical prediction model according to predicted electricity consumption data before a prediction period and real data of the prediction period; the loss function of the history prediction model is determined according to the predicted electricity consumption data before the prediction period, the real data in the prediction period, the electricity consumption behavior characteristics of the target user and the weight information of the target user in the user group; the target users can be users meeting preset electricity consumption and data acquisition conditions in the user group and are used for representing electricity consumption characteristics of the whole user group, and the history prediction model is trained through electricity consumption data acquired from the selected target users, so that model training effect is ensured, and model training efficiency is improved. Further, considering the influence of the selected target user on model training, the difference between the predicted electricity consumption data before the prediction period and the predicted real data can be corrected according to the electricity consumption behavior characteristics of the target user and the weight of the target user occupying the user group, and a loss function is calculated according to the corrected difference, so that the electricity consumption prediction model is optimized. The electricity consumption behavior characteristics of the target user and the predicted electricity consumption data of the target user in the prediction period, which occupy the user group, are output by the history prediction model; the prediction period real data are real electricity utilization data of the target user in the prediction period; the historical electricity consumption data of the user to be predicted is input into the electricity consumption prediction model to obtain the predicted electricity consumption data of the user to be predicted, the predicted electricity consumption data is different from the whole steady state data selected from the power grid in the prior art to conduct electricity consumption prediction, the efficiency is low, the influence of the electricity consumption behavior of the user on a model prediction result is ignored, the power grid user is divided into a plurality of user groups according to user attributes, each user group corresponds to a specific electricity consumption prediction model, and the result of model training is corrected according to the electricity consumption behavior characteristics of a target user selected in the user group and the relation between the power consumption behavior characteristics and the whole group, so that the efficiency and the accuracy of electricity consumption prediction are improved.

Fig. 3 is a schematic structural diagram of an electricity consumption prediction device according to an embodiment of the present invention, and the specific embodiment of the present invention is not limited to the specific implementation of the electricity consumption prediction device.

As shown in fig. 3, the electricity consumption amount prediction apparatus may include: a processor 402, a communication interface (Communications Interface) 404, a memory 406, and a communication bus 408.

Wherein: processor 402, communication interface 404, and memory 406 communicate with each other via communication bus 408. A communication interface 404 for communicating with network elements of other devices, such as clients or other servers. The processor 402 is configured to execute the program 410, and may specifically perform the relevant steps in the embodiment of the power consumption prediction method described above.

In particular, program 410 may include program code including computer-executable instructions.

The processor 402 may be a central processing unit CPU, or a specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in the power consumption prediction device may be the same type of processor, such as one or more CPUs; but may also be different types of processors such as one or more CPUs and one or more ASICs.

Memory 406 for storing programs 410. Memory 406 may comprise high-speed RAM memory or may also include non-volatile memory (non-volatile memory), such as at least one disk memory.

Program 410 may be specifically invoked by processor 402 to use the power prediction device to:

acquiring a power consumption prediction model corresponding to a user group where a user to be predicted is located; the electricity consumption prediction model is obtained by training a historical prediction model according to predicted electricity consumption data before a prediction period and real data of the prediction period; the loss function of the history prediction model is determined according to the predicted electricity consumption data before the prediction period, the real data in the prediction period, the electricity consumption behavior characteristics of the target user and the weight information of the target user in the user group; the predicted electricity consumption data before the prediction period is the predicted electricity consumption data of the target user in the prediction period, which is output by the history prediction model; the prediction period real data are real electricity utilization data of the target user in the prediction period;

and inputting the historical electricity consumption data of the user to be predicted into the electricity consumption prediction model to obtain the predicted electricity consumption data of the user to be predicted.

The execution process of the electricity consumption prediction device provided by the embodiment of the present invention is substantially the same as that of the foregoing method embodiment, and will not be described again.

The electricity consumption prediction device provided by the embodiment of the invention obtains the electricity consumption prediction model corresponding to the user group where the user to be predicted is located; the electricity consumption prediction model is obtained by training a historical prediction model according to predicted electricity consumption data before a prediction period and real data of the prediction period; the loss function of the history prediction model is determined according to the predicted electricity consumption data before the prediction period, the real data in the prediction period, the electricity consumption behavior characteristics of the target user and the weight information of the target user in the user group; the target users can be users meeting preset electricity consumption and data acquisition conditions in the user group and are used for representing electricity consumption characteristics of the whole user group, and the history prediction model is trained through electricity consumption data acquired from the selected target users, so that model training effect is ensured, and model training efficiency is improved. Further, considering the influence of the selected target user on model training, the difference between the predicted electricity consumption data before the prediction period and the predicted real data can be corrected according to the electricity consumption behavior characteristics of the target user and the weight of the target user occupying the user group, and a loss function is calculated according to the corrected difference, so that the electricity consumption prediction model is optimized. The electricity consumption behavior characteristics of the target user and the predicted electricity consumption data of the target user in the prediction period, which occupy the user group, are output by the history prediction model; the prediction period real data are real electricity utilization data of the target user in the prediction period; the historical electricity consumption data of the user to be predicted is input into the electricity consumption prediction model to obtain the predicted electricity consumption data of the user to be predicted, the predicted electricity consumption data is different from the whole steady state data selected from the power grid in the prior art to conduct electricity consumption prediction, the efficiency is low, the influence of the electricity consumption behavior of the user on a model prediction result is ignored, the power grid user is divided into a plurality of user groups according to user attributes, each user group corresponds to a specific electricity consumption prediction model, and the result of model training is corrected according to the electricity consumption behavior characteristics of a target user selected in the user group and the relation between the power consumption behavior characteristics and the whole group, so that the efficiency and the accuracy of electricity consumption prediction are improved.

The embodiment of the invention provides a computer readable storage medium, which stores at least one executable instruction, and when the executable instruction runs on electricity consumption prediction equipment, the electricity consumption prediction equipment executes the electricity consumption prediction method in any method embodiment.

The executable instructions may be specifically configured to cause the power consumption prediction device to:

acquiring a power consumption prediction model corresponding to a user group where a user to be predicted is located; the electricity consumption prediction model is obtained by training a historical prediction model according to predicted electricity consumption data before a prediction period and real data of the prediction period; the loss function of the history prediction model is determined according to the predicted electricity consumption data before the prediction period, the real data in the prediction period, the electricity consumption behavior characteristics of the target user and the weight information of the target user in the user group; the predicted electricity consumption data before the prediction period is the predicted electricity consumption data of the target user in the prediction period, which is output by the history prediction model; the prediction period real data are real electricity utilization data of the target user in the prediction period;

and inputting the historical electricity consumption data of the user to be predicted into the electricity consumption prediction model to obtain the predicted electricity consumption data of the user to be predicted.

The execution process of the executable instructions stored in the computer readable storage medium provided by the embodiment of the present invention is substantially the same as that of the foregoing method embodiment, and will not be repeated.

The executable instructions stored in the computer readable storage medium provided by the embodiment of the invention are used for obtaining the power consumption prediction model corresponding to the user group where the user to be predicted is located; the electricity consumption prediction model is obtained by training a historical prediction model according to predicted electricity consumption data before a prediction period and real data of the prediction period; the loss function of the history prediction model is determined according to the predicted electricity consumption data before the prediction period, the real data in the prediction period, the electricity consumption behavior characteristics of the target user and the weight information of the target user in the user group; the target users can be users meeting preset electricity consumption and data acquisition conditions in the user group and are used for representing electricity consumption characteristics of the whole user group, and the history prediction model is trained through electricity consumption data acquired from the selected target users, so that model training effect is ensured, and model training efficiency is improved. Further, considering the influence of the selected target user on model training, the difference between the predicted electricity consumption data before the prediction period and the predicted real data can be corrected according to the electricity consumption behavior characteristics of the target user and the weight of the target user occupying the user group, and a loss function is calculated according to the corrected difference, so that the electricity consumption prediction model is optimized. The electricity consumption behavior characteristics of the target user and the predicted electricity consumption data of the target user in the prediction period, which occupy the user group, are output by the history prediction model; the prediction period real data are real electricity utilization data of the target user in the prediction period; the historical electricity consumption data of the user to be predicted is input into the electricity consumption prediction model to obtain the predicted electricity consumption data of the user to be predicted, the predicted electricity consumption data is different from the whole steady state data selected from the power grid in the prior art to conduct electricity consumption prediction, the efficiency is low, the influence of the electricity consumption behavior of the user on a model prediction result is ignored, the power grid user is divided into a plurality of user groups according to user attributes, each user group corresponds to a specific electricity consumption prediction model, and the result of model training is corrected according to the electricity consumption behavior characteristics of a target user selected in the user group and the relation between the power consumption behavior characteristics and the whole group, so that the efficiency and the accuracy of electricity consumption prediction are improved.

The embodiment of the invention provides a power consumption prediction device which is used for executing the power consumption prediction method.

Embodiments of the present invention provide a computer program that may be invoked by a processor to perform the power usage prediction method of any of the method embodiments described above using a power usage prediction device.

An embodiment of the present invention provides a computer program product comprising a computer program stored on a computer readable storage medium, the computer program comprising program instructions which, when run on a computer, cause the computer to perform the method for predicting power usage in any of the method embodiments described above.

The algorithms or displays presented herein are not inherently related to any particular computer, virtual system, or other apparatus. Various general-purpose systems may also be used with the teachings herein. The required structure for a construction of such a system is apparent from the description above. In addition, embodiments of the present invention are not directed to any particular programming language. It will be appreciated that the teachings of the present invention described herein may be implemented in a variety of programming languages, and the above description of specific languages is provided for disclosure of enablement and best mode of the present invention.

In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.

Similarly, it should be appreciated that in the above description of exemplary embodiments of the invention, various features of the embodiments of the invention are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure and aiding in the understanding of one or more of the various inventive aspects. However, the disclosed method should not be construed as reflecting the intention that: i.e., the claimed invention requires more features than are expressly recited in each claim.

Those skilled in the art will appreciate that the modules in the apparatus of the embodiments may be adaptively changed and disposed in one or more apparatuses different from the embodiments. The modules or units or components of the embodiments may be combined into one module or unit or component, and they may be divided into a plurality of sub-modules or sub-units or sub-components. Any combination of all features disclosed in this specification (including any accompanying claims, abstract and drawings), and all of the processes or units of any method or apparatus so disclosed, may be used in combination, except insofar as at least some of such features and/or processes or units are mutually exclusive. Each feature disclosed in this specification (including any accompanying claims, abstract and drawings), may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.

It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, and that those skilled in the art will be able to design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention may be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by one and the same item of hardware. The use of the words first, second, third, etc. do not denote any order. These words may be interpreted as names. The steps in the above embodiments should not be construed as limiting the order of execution unless specifically stated.

Claims (10)

1. A method of predicting power usage, the method comprising:

acquiring a power consumption prediction model corresponding to a user group where a user to be predicted is located; the electricity consumption prediction model is obtained by training a historical prediction model according to predicted electricity consumption data before a prediction period and real data of the prediction period; the loss function of the history prediction model is determined according to the predicted electricity consumption data before the prediction period, the real data in the prediction period, the electricity consumption behavior characteristics of the target user and the weight information of the target user in the user group; the predicted electricity consumption data before the prediction period is the predicted electricity consumption data of the target user in the prediction period, which is output by the history prediction model; the prediction period real data are real electricity utilization data of the target user in the prediction period;

And inputting the historical electricity consumption data of the user to be predicted into the electricity consumption prediction model to obtain the predicted electricity consumption data of the user to be predicted.

2. The method according to claim 1, further comprising, prior to the obtaining the power consumption prediction model corresponding to the user group in which the user to be predicted is located:

determining cross entropy between the predicted electricity consumption data before the prediction period and the real data in the prediction period;

determining cross entropy correction parameters corresponding to the user group according to the electricity behavior characteristics and the weight information;

and determining the loss function according to the cross entropy, the cross entropy correction parameter and a preset cross entropy threshold.

3. The method of claim 2, wherein the cross-entropy correction parameter comprises a first correction parameter; the determining the cross entropy correction parameter corresponding to the user group according to the electricity behavior feature and the weight information further includes:

determining the acquired data extreme value and the mean value information of the target user according to the electricity behavior characteristics;

and determining the first correction parameter according to the acquired data extreme value and the mean value information.

4. The method of claim 2, wherein the cross-entropy correction parameter comprises a second correction parameter; the target users comprise users with electricity consumption and data acquisition capacity meeting preset conditions in the user group; the determining the cross entropy correction parameter corresponding to the user group according to the electricity behavior feature and the weight information further includes:

determining the weight information according to the proportion of the target user to the total number of users in the user group;

and determining the second correction parameter according to the weight information.

5. The method of claim 2, wherein the determining the loss function based on the cross entropy, cross entropy modification parameters, and a preset cross entropy threshold further comprises:

correcting the cross entropy according to the cross entropy correction parameter to obtain corrected cross entropy;

and determining the loss function according to the comparison result of the corrected cross entropy and the cross entropy threshold.

6. The method of claim 1, wherein the group of users is one of a plurality of selectable groups; the plurality of selectable groups are obtained by dividing all users in the target area according to the user characteristic images; the history prediction model comprises an updated global model; before the power consumption prediction model corresponding to the user group where the user to be predicted is located is obtained, the method further comprises:

The initial prediction model is issued to the edge side server corresponding to each optional group;

for each optional group, training the initial prediction model through the edge side server according to the predicted electricity consumption data before the prediction period and the predicted real data in the prediction period corresponding to the optional group to obtain an updated group model corresponding to the optional group;

the updated group models returned by all the edge side servers are aggregated according to the group characteristic information of the selectable groups, so that updated global models are obtained;

and sending the updated global model to each edge server so that the edge server carries out the next training on the updated global model.

7. The method of claim 6, wherein aggregating the updated group models returned by all the edge side servers according to the group feature information of the optional group, to obtain the updated global model, further comprises:

determining group data reference weights corresponding to the optional groups and similarity among the optional groups according to the group characteristic information;

And aggregating the updated group model according to the similarity and the group data reference weight to obtain the updated global model.

8. A power consumption prediction apparatus, characterized in that the apparatus comprises:

the acquisition module is used for acquiring a power consumption prediction model corresponding to a user group where a user to be predicted is located; the electricity consumption prediction model is obtained by training a historical prediction model according to predicted electricity consumption data before a prediction period and real data of the prediction period; the loss function of the history prediction model is determined according to the predicted electricity consumption data before the prediction period, the real data in the prediction period, the electricity consumption behavior characteristics of the target user and the weight information of the target user relative to the user group; the predicted electricity consumption data before the prediction period is the predicted electricity consumption data of the target user in the prediction period, which is output by the history prediction model; the prediction period real data are real electricity utilization data of the target user in the prediction period;

and the prediction module is used for inputting the historical electricity consumption data of the user to be predicted into the electricity consumption prediction model to obtain the predicted electricity consumption data of the user to be predicted.

9. A power consumption amount prediction apparatus, characterized by comprising: the device comprises a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface complete communication with each other through the communication bus;

the memory is configured to store at least one executable instruction that causes the processor to perform the operations of the power consumption prediction method according to any one of claims 1-7.

10. A computer readable storage medium, wherein at least one executable instruction is stored in the storage medium, which when run on a power consumption prediction device, causes the power consumption prediction device to perform the operations of the power consumption prediction method according to any one of claims 1-7.