A method for monitoring electricity theft in charging piles based on big data

Abstract

The present invention discloses a charging pile electricity theft behavior monitoring method based on big data, which belongs to the field of big data technology and electricity theft monitoring technology. By acquiring historical big data and artificial labels, and learning the historical big data and artificial labels, a rule for intelligently detecting the electricity theft behavior of charging piles is formed. Finally, real-time data is identified through the rule for intelligently detecting the electricity theft behavior of charging piles, thereby realizing real-time automatic monitoring of the electricity theft behavior of charging piles, and improving the monitoring efficiency and identification accuracy of the electricity theft of charging piles.

Description

Technical Field

[0001] The present invention belongs to the field of big data technology and electricity theft monitoring technology, and specifically relates to a charging pile electricity theft behavior monitoring method based on big data. Background Art

[0002] An electric vehicle (EV) is a means of transportation that uses electricity as its main power source. It is driven by an electric motor instead of a traditional internal combustion engine. The operating principle of an electric vehicle is relatively simple. The power comes from the electric motor, which converts the electrical energy stored in the battery into mechanical energy to propel the vehicle forward. Electric vehicles do not require fuel, so they do not emit exhaust gas during operation, making them an effective way to achieve green travel. With the widespread use of electric vehicles, the demand for charging piles continues to increase. However, the phenomenon of electricity theft in charging piles occurs from time to time, causing economic losses to charging pile operators and users. Traditional methods for monitoring electricity theft in charging piles mainly rely on manual inspections and real-time alarm systems, with low monitoring efficiency, making it difficult to achieve real-time and accurate identification of electricity theft in charging piles. Summary of the invention

[0003] The present invention provides a charging pile electricity theft monitoring method based on big data, which is used to solve the problem that the traditional charging pile electricity theft monitoring method mainly relies on manual inspection and real-time alarm system, has low monitoring efficiency, and is difficult to achieve real-time and accurate identification of charging pile electricity theft behavior.

[0004] A method for monitoring electricity theft behavior of charging piles based on big data, comprising:

[0005] Obtaining historical electricity consumption big data corresponding to the charging pile and manual labels corresponding to the historical electricity consumption data, and cleaning the historical electricity consumption big data and manual labels corresponding to the historical electricity consumption data to obtain the cleaned historical electricity consumption big data and manual labels corresponding to the historical electricity consumption data;

[0006] A deep learning model is used to build a charging pile electricity theft detection model, and the charging pile electricity theft detection model is optimized using the cleaned historical electricity consumption big data and the manual labels corresponding to the historical electricity consumption data to obtain the optimized charging pile electricity theft detection model;

[0007] The optimized charging pile electricity theft behavior detection model is deployed on the electricity consumption data monitoring platform, and the real-time electricity consumption data is analyzed through the deployed charging pile electricity theft behavior detection model to determine the monitoring results of the charging pile electricity theft behavior.

[0008] Furthermore, the historical electricity consumption big data corresponding to the charging pile includes electricity consumption type data and charging pile electric energy meter information;

[0009] The artificial labels corresponding to the historical electricity consumption data include whether there is electricity theft at the charging pile or there is no electricity theft at the charging pile;

[0010] Among them, the electricity consumption type data includes the time series data of daily electricity consumption of charging piles and the time series data of daily line loss in the substation area;

[0011] The charging pile electric energy meter information includes the electric energy meter current imbalance event record, the electric energy meter overcurrent event record, the electric energy meter constant magnetic field interference event record, the electric energy meter cover opening event record, the electric energy meter undervoltage event record, the electric energy meter loss of current event record, the electric energy meter loss of pressure event record and the electric energy meter reverse running event record.

[0012] Furthermore, the historical electricity consumption big data and the manual labels corresponding to the historical electricity consumption data are cleaned to obtain the cleaned historical electricity consumption big data and the manual labels corresponding to the historical electricity consumption data, including:

[0013] The historical electricity consumption big data with abnormal values ​​and / or missing values ​​and the artificial labels corresponding to the historical electricity consumption data are deleted to obtain the cleaned historical electricity consumption big data and the artificial labels corresponding to the historical electricity consumption data.

[0014] Furthermore, a deep learning model is used to construct a charging pile electricity theft behavior detection model, including: using a convolutional neural network to construct a charging pile electricity theft behavior detection model.

[0015] Furthermore, the charging pile electricity theft behavior detection model is optimized by using the cleaned historical electricity consumption big data and the artificial labels corresponding to the historical electricity consumption data, and the optimized charging pile electricity theft behavior detection model is obtained, including:

[0016] The chaotic mapping initialization method is used to initialize the hyperparameters of the charging pile electricity theft detection model multiple times to obtain multiple different hyperparameter encodings.

[0017] According to the cleaned historical electricity consumption big data and the manual labels corresponding to the historical electricity consumption data, the error function value of each hyperparameter encoding is obtained;

[0018] Determine the optimal hyperparameter encoding according to the error function value of each hyperparameter encoding;

[0019] All hyperparameter encodings are randomly divided into two sub-populations;

[0020] The first information interaction strategy is adopted to make the hyperparameter encodings in the two sub-populations perform initial information interaction to obtain the sub-populations after initial update;

[0021] According to the optimal hyperparameter encoding, a second information interaction strategy is adopted to make the hyperparameter encodings in the two sub-populations perform secondary information interaction to obtain a sub-population after secondary update;

[0022] According to the optimal hyperparameter encoding, the third information interaction strategy is adopted to make the hyperparameter encodings in the two sub-populations interact with each other three times, and the sub-populations are obtained after three updates;

[0023] The two sub-populations after three updates are merged into a complete population, and the perturbation update is performed on each hyperparameter encoding in the complete population to obtain the updated complete population;

[0024] Based on the updated complete population, the optimal hyperparameter encoding is re-obtained, and it is determined whether the current optimization times have reached the maximum optimization times. If so, the optimal hyperparameter encoding is used as the final hyperparameter of the charging pile electricity theft behavior detection model to obtain the optimized charging pile electricity theft behavior detection model. Otherwise, the step of obtaining the error function value of each hyperparameter encoding is returned.

[0025] Furthermore, according to the cleaned historical electricity consumption big data and the manual labels corresponding to the historical electricity consumption data, the error function value encoded by each hyperparameter is obtained, including:

[0026] After applying the hyperparameters included in the hyperparameter encoding to the charging pile electricity theft detection model, the cleaned historical electricity consumption big data is used as the actual input of the charging pile electricity theft detection model, and the manual labels corresponding to the historical electricity consumption data are used as the expected output of the charging pile electricity theft detection model. The mean square error or root mean square error corresponding to the charging pile electricity theft detection model is obtained, and the error function value of the hyperparameter encoding is obtained;

[0027] Traverse all hyperparameter encodings and obtain the error function value of each hyperparameter encoding.

[0028] Furthermore, the first information interaction strategy is adopted to make the hyperparameter encodings in the two sub-populations perform initial information interaction to obtain the sub-populations after initial update, including:

[0029] Pair the hyperparameter encodings in the two subpopulations one by one;

[0030] For any pair of hyperparameter encodings, determine the hyperparameter encoding with the larger error function value And the hyperparameter encoding with a small error function value Hyperparameter encoding Encoding hyperparameters Paired individuals;

[0031] Hyperparameter encoding for small error function values Directly encode it as the updated hyperparameter;

[0032] Hyperparameter encoding for large error function values Encoding it with hyperparameters Perform information exchange to obtain updated hyperparameter encoding for:

[0033]

[0034]

[0035] in, Represents hyperparameter encoding The update amount in the t+1th optimization process, Represents the updated hyperparameter encoding Represents hyperparameter encoding The update amount in the tth optimization process. When there is no update amount in the tth optimization process, then is 0; ω represents the inertia weight, r1 represents a random number uniformly distributed in [0,1], and r2 represents a random number uniformly distributed in [0,1]. represents the interaction control factor, Represents hyperparameter encoding The average code corresponding to the subpopulation;

[0036] Traverse each pair of hyperparameter encodings to obtain the sub-population after the initial update.

[0037] Furthermore, according to the optimal hyperparameter encoding, a second information interaction strategy is adopted to make the hyperparameter encodings in the two subpopulations perform secondary information interaction to obtain a subpopulation after secondary update, including:

[0038] Pair the hyperparameter encodings in the two subpopulations one by one;

[0039] For any pair of hyperparameter encodings, determine the hyperparameter encoding with the larger error function value And the hyperparameter encoding with a small error function value Hyperparameter encoding Encoding hyperparameters Paired individuals;

[0040] Hyperparameter encoding for small error function values Directly encode it as the updated hyperparameter;

[0041] Hyperparameter encoding for large error function values Encoding it with hyperparameters Perform information exchange to obtain updated hyperparameter encoding for:

[0042]

[0043]

[0044] in, Represents hyperparameter encoding The update amount in the t+1th optimization process, Represents the updated hyperparameter encoding Represents hyperparameter encoding The update amount in the tth optimization process. When there is no update amount in the tth optimization process, then is 0; ω represents the inertia weight, r1 represents a random number uniformly distributed in [0,1], and r2 represents a random number uniformly distributed in [0,1]. represents the interaction control factor, Represents hyperparameter encoding The average code corresponding to the subpopulation, r3 represents a random number uniformly distributed in [0,1], represents the optimal hyperparameter encoding;

[0045] Traverse each pair of hyperparameter encodings to obtain the sub-population after secondary update.

[0046] Furthermore, according to the optimal hyperparameter encoding, the third information interaction strategy is adopted to make the hyperparameter encodings in the two subpopulations perform information interaction three times, and obtain three updated subpopulations, including:

[0047] Determine the current optimization times t, and determine the decision factor as t / T according to the current optimization times t; where T represents the maximum optimization times;

[0048] For any pair of hyperparameter encodings, when the decision factor is less than or equal to 0.5, the hyperparameter encoding is updated as follows:

[0049]

[0050]

[0051] in, and Paired hyperparameter encodings, Represents the updated hyperparameter encoding Represents the new hyperparameter encoding r4 represents a random number between (0,1), represents the optimal hyperparameter encoding, bf represents the information interaction factor;

[0052] For any pair of hyperparameter encodings, when the decision factor is greater than 0.5, the hyperparameter encoding is updated as follows:

[0053]

[0054]

[0055] Traverse each pair of hyperparameter encodings and obtain three updated sub-populations.

[0056] Furthermore, the two sub-populations after three updates are merged into a complete population, and a perturbation update is performed on each hyperparameter encoding in the complete population to obtain an updated complete population, including:

[0057] Merge the two sub-populations after three updates into a complete population;

[0058] The perturbation update for the hyperparameter encoding in the complete population is:

[0059]

[0060] in, represents the kth hyperparameter encoding in the complete population, Indicates updated k=1,2,…,K, K represents the total number of hyperparameter encodings in the complete population, π represents pi, t represents the current optimization number, T represents the maximum optimization number, and r5 represents a random number between (0,1).

[0061] The present invention provides a charging pile electricity theft behavior monitoring method based on big data, which obtains historical big data and artificial labels, and learns the historical big data and artificial labels to form rules for intelligent detection of charging pile electricity theft behavior. Finally, real-time data is identified through the rules for intelligent detection of charging pile electricity theft behavior, thereby realizing real-time automatic monitoring of charging pile electricity theft behavior, and improving the charging pile electricity theft monitoring efficiency and identification accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0063] Figure 1 A flowchart of a method for monitoring electricity theft in a charging pile based on big data is provided in an embodiment of the present invention.

[0064] The above drawings have shown clear embodiments of the present invention, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present invention in any way, but to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0065] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0066] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0067] like Figure 1 As shown, a method for monitoring electricity theft behavior of a charging pile based on big data in an embodiment of the present invention includes:

[0068] S1. Obtain historical electricity consumption big data corresponding to the charging pile and artificial labels corresponding to the historical electricity consumption data, clean the historical electricity consumption big data and artificial labels corresponding to the historical electricity consumption data, and obtain the cleaned historical electricity consumption big data and artificial labels corresponding to the historical electricity consumption data.

[0069] Artificial labels can be determined based on historical electricity consumption data, or various electricity theft behaviors can be artificially created. At the same time, historical electricity consumption data can be collected to ensure that the electricity consumption data can be accurately identified in the end. After the data is collected, some data may be missing or abnormal during the transmission process, so the data needs to be cleaned. For example, if a certain data has a negative value when it is obviously impossible to be negative, then it should be abnormal data.

[0070] It is worth mentioning that in addition to cleaning data using preset rules, data can also be cleaned manually.

[0071] S2. Use a deep learning model to build a charging pile electricity theft detection model, and use the cleaned historical electricity consumption big data and the manual labels corresponding to the historical electricity consumption data to optimize the charging pile electricity theft detection model to obtain the optimized charging pile electricity theft detection model.

[0072] The process of using deep learning models to identify electricity consumption data is essentially a data classification process, so a convolutional neural network can be used to build a charging pile electricity theft detection model to analyze electricity consumption data. However, the deep learning model cannot be used directly after it is built, so it is necessary to optimize the hyperparameters of the charging pile electricity theft detection model to obtain the optimized charging pile electricity theft detection model.

[0073] S3. Deploy the optimized charging pile electricity theft behavior detection model on the electricity consumption data monitoring platform, and analyze the real-time electricity consumption data through the deployed charging pile electricity theft behavior detection model to determine the monitoring results of the charging pile electricity theft behavior.

[0074] After the optimization of the charging pile electricity theft detection model is completed, it can be deployed on the electricity consumption data monitoring platform (i.e. the management platform or server of the charging pile merchant) to receive real-time electricity consumption data and input it into the model for identification, thereby realizing effective mining of electricity consumption data and determining the monitoring results of electricity theft.

[0075] Optionally, different charging pile data may be identified separately, so that after electricity theft occurs, the abnormal charging pile can be quickly located.

[0076] In an embodiment of the present invention, the historical electricity consumption big data corresponding to the charging pile includes electricity consumption type data and charging pile electric energy meter information.

[0077] The artificial labels corresponding to the historical electricity usage data include whether there is electricity theft at the charging pile or whether there is no electricity theft at the charging pile.

[0078] Among them, the electricity consumption type data includes the time series data of daily electricity consumption of charging piles and the time series data of daily line losses in the substation area.

[0079] The charging pile electric energy meter information includes the electric energy meter current imbalance event record, the electric energy meter overcurrent event record, the electric energy meter constant magnetic field interference event record, the electric energy meter cover opening event record, the electric energy meter undervoltage event record, the electric energy meter loss of current event record, the electric energy meter loss of pressure event record and the electric energy meter reverse running event record.

[0080] It is worth noting that the above-mentioned electricity consumption big data is only a preferred choice for the embodiment of the present invention. Other types of data can also be used for training and identification, so that different analysis effects can be identified. In order to accurately identify real-time electricity consumption data, it is necessary to ensure that the data type and data length of historical electricity consumption big data are the same as those of real-time electricity consumption data. Therefore, the maximum length of each type of parameter can be used as the standard length. When the length of the same type of parameter is less than the standard length, the method of padding the data with zeros is used to fill it to the standard length. For example, when the maximum electricity consumption in a certain electricity consumption data is a three-digit number, it is necessary to pad zeros before the electricity consumption less than three digits, so that all sequences can have the same length.

[0081] In an embodiment of the present invention, cleaning the historical electricity consumption big data and the artificial labels corresponding to the historical electricity consumption data to obtain the cleaned historical electricity consumption big data and the artificial labels corresponding to the historical electricity consumption data includes:

[0082] The historical electricity consumption big data with abnormal values ​​and / or missing values ​​and the artificial labels corresponding to the historical electricity consumption data are deleted to obtain the cleaned historical electricity consumption big data and the artificial labels corresponding to the historical electricity consumption data.

[0083] It is worth noting that the cleaning rules of the embodiment of the present invention are relatively simple, and other data cleaning rules can also be used to clean the historical electricity consumption big data, thereby ensuring the accuracy of the electricity consumption data.

[0084] In an embodiment of the present invention, a deep learning model is used to construct a charging pile electricity theft detection model, including: using a convolutional neural network to construct a charging pile electricity theft detection model. It is worth noting that the convolutional neural network is only a preferred choice for this embodiment, and other classification neural networks can also be used to construct a charging pile electricity theft detection model.

[0085] Optionally, when the historical electricity consumption big data has many data features, a data dimensionality reduction method can be used to reduce its dimensionality, thereby ensuring that the charging pile electricity theft behavior detection model can identify the electricity consumption data. After the historical electricity consumption big data is reduced in dimensionality, the real-time electricity consumption data also needs to be reduced in dimensionality.

[0086] In the prior art, optimization algorithms such as particle swarm optimization are often used to optimize deep learning models so that the deep learning models can analyze data. Although the particle swarm algorithm has good convergence speed and convergence accuracy, its global search ability is poor and it is easy to fall into local optimality, resulting in poor analysis ability of the deep learning model. Therefore, an embodiment of the present invention provides an optimization method for a deep learning model to solve the technical problems existing in the current technology.

[0087] In an embodiment of the present invention, the charging pile electricity theft behavior detection model is optimized by using the cleaned historical electricity consumption big data and the artificial labels corresponding to the historical electricity consumption data, and the optimized charging pile electricity theft behavior detection model is obtained, including:

[0088] The chaotic mapping initialization method is used to initialize the hyperparameters of the charging pile electricity theft detection model multiple times to obtain multiple different hyperparameter encodings.

[0089] Through the chaos mapping initialization method, the initial solution can be distributed more evenly in the solution space, which helps to improve the overall convergence speed and convergence effect of the algorithm.

[0090] According to the cleaned historical electricity consumption big data and the manual labels corresponding to the historical electricity consumption data, the error function value encoded by each hyperparameter is obtained.

[0091] Optionally, a root mean square error function, a mean square error function or a cross entropy error function may be used to obtain the error function value of each hyperparameter encoding. It is worth noting that the above error function is only used as an example, and other error functions may also be used to obtain the error function value.

[0092] According to the error function value of each hyperparameter encoding, the optimal hyperparameter encoding is determined, that is, the hyperparameter encoding with the smallest error function value is taken as the optimal hyperparameter encoding.

[0093] All hyperparameter encodings are randomly divided into two sub-populations.

[0094] The first information interaction strategy is adopted to make the hyperparameter encodings in the two sub-populations perform initial information interaction to obtain the sub-populations after the initial update.

[0095] According to the optimal hyperparameter encoding, a second information interaction strategy is adopted to make the hyperparameter encodings in the two sub-populations perform secondary information interaction to obtain a sub-population after secondary update.

[0096] According to the optimal hyperparameter encoding, the third information interaction strategy is adopted to make the hyperparameter encodings in the two sub-populations interact with each other three times, and the sub-populations are obtained after three updates.

[0097] The two sub-populations after three updates are merged into a complete population, and a perturbation update is performed on each hyperparameter encoding in the complete population to obtain an updated complete population.

[0098] Based on the updated complete population, the optimal hyperparameter encoding is re-obtained, and it is determined whether the current optimization times have reached the maximum optimization times. If so, the optimal hyperparameter encoding is used as the final hyperparameter of the charging pile electricity theft behavior detection model to obtain the optimized charging pile electricity theft behavior detection model. Otherwise, the step of obtaining the error function value of each hyperparameter encoding is returned.

[0099] The algorithm provided by the embodiment of the present invention can effectively perform global optimization and local optimization through dual-group information interaction, thereby improving the training effect of the deep learning model.

[0100] In the embodiment of the present invention, according to the cleaned historical electricity consumption big data and the manual labels corresponding to the historical electricity consumption data, the error function value of each hyperparameter encoding is obtained, including:

[0101] After applying the hyperparameters contained in the hyperparameter encoding to the charging pile electricity theft detection model, the cleaned historical electricity consumption big data is used as the actual input of the charging pile electricity theft detection model, and the manual labels corresponding to the historical electricity consumption data are used as the expected output of the charging pile electricity theft detection model. The mean square error or root mean square error corresponding to the charging pile electricity theft detection model is obtained, and the error function value of the hyperparameter encoding is obtained.

[0102] Traverse all hyperparameter encodings and obtain the error function value of each hyperparameter encoding.

[0103] In the embodiment of the present invention, the first information interaction strategy is adopted to make the hyperparameter encodings in the two sub-populations perform initial information interaction to obtain the initially updated sub-populations, including:

[0104] The hyperparameter encodings in the two subpopulations are paired one to one.

[0105] For any pair of hyperparameter encodings, determine the hyperparameter encoding with the larger error function value And the hyperparameter encoding with a small error function value Hyperparameter encoding Encoding hyperparameters paired individuals.

[0106] Hyperparameter encoding for small error function values Directly encode it as the updated hyperparameter.

[0107] Hyperparameter encoding for large error function values Encoding it with hyperparameters Perform information exchange to obtain updated hyperparameter encoding for:

[0108]

[0109]

[0110] in, Represents hyperparameter encoding The update amount in the t+1th optimization process, Represents the updated hyperparameter encoding Represents hyperparameter encoding The update amount in the tth optimization process. When there is no update amount in the tth optimization process, then is 0. ω represents the inertia weight, r1 represents a random number uniformly distributed in [0,1], and r2 represents a random number uniformly distributed in [0,1]. represents the interaction control factor, Represents hyperparameter encoding The average code corresponding to the subpopulation.

[0111] Traverse each pair of hyperparameter encodings to obtain the sub-population after the initial update.

[0112] The first information interaction strategy provided by the embodiment of the present invention can effectively utilize individual information interaction, individual historical information and group information interaction, so that a certain global optimization can be performed while effectively exploring unknown areas.

[0113] In the embodiment of the present invention, according to the optimal hyperparameter encoding, a second information interaction strategy is adopted to make the hyperparameter encodings in the two subpopulations perform secondary information interaction to obtain a subpopulation after secondary update, including:

[0114] The hyperparameter encodings in the two subpopulations are paired one to one.

[0115] For any pair of hyperparameter encodings, determine the hyperparameter encoding with the larger error function value And the hyperparameter encoding with a small error function value Hyperparameter encoding Encoding hyperparameters paired individuals.

[0116] Hyperparameter encoding for small error function values Directly encode it as the updated hyperparameter.

[0117] Hyperparameter encoding for large error function values Encoding it with hyperparameters Perform information exchange to obtain updated hyperparameter encoding for:

[0118]

[0119]

[0120] in, Represents hyperparameter encoding The update amount in the t+1th optimization process, Represents the updated hyperparameter encoding Represents hyperparameter encoding The update amount in the tth optimization process. When there is no update amount in the tth optimization process, then is 0. ω represents the inertia weight, r1 represents a random number uniformly distributed in [0,1], and r2 represents a random number uniformly distributed in [0,1]. represents the interaction control factor, Represents hyperparameter encoding The average code corresponding to the subpopulation, r3 represents a random number uniformly distributed in [0,1], represents the optimal hyperparameter encoding.

[0121] Traverse each pair of hyperparameter encodings to obtain the sub-population after secondary update.

[0122] The second information interaction strategy provided by the embodiment of the present invention introduces optimal hyperparameter encoding, which can effectively increase the local convergence effect, thereby improving the overall convergence accuracy of the algorithm.

[0123] In the embodiment of the present invention, according to the optimal hyperparameter encoding, the third information interaction strategy is adopted to make the hyperparameter encodings in the two subpopulations perform information interaction three times, and obtain three updated subpopulations, including:

[0124] Determine the current optimization times t, and determine the decision factor as t / T according to the current optimization times t, where T represents the maximum optimization times.

[0125] For any pair of hyperparameter encodings, when the decision factor is less than or equal to 0.5, the hyperparameter encoding is updated as follows:

[0126]

[0127]

[0128] in, and Paired hyperparameter encodings, Represents the updated hyperparameter encoding Represents the new hyperparameter encoding r4 represents a random number between (0,1), represents the optimal hyperparameter encoding, and bf represents the information interaction factor.

[0129] For any pair of hyperparameter encodings, when the decision factor is greater than 0.5, the hyperparameter encoding is updated as follows:

[0130]

[0131]

[0132] Traverse each pair of hyperparameter encodings and obtain three updated sub-populations.

[0133] The third information interaction strategy provided by the embodiment of the present invention can utilize individual information interaction to perform optimization in the early stage of the algorithm, and only interact with the optimal hyperparameter encoding in the later stage of the algorithm, thereby improving the convergence accuracy in the later stage of the algorithm and ensuring the optimization effect of the algorithm.

[0134] In the embodiment of the present invention, the two sub-populations after three updates are merged into a complete population, and a perturbation update is performed on each hyperparameter encoding in the complete population to obtain an updated complete population, including:

[0135] The two subpopulations after three updates are merged into a complete population.

[0136] The perturbation update for the hyperparameter encoding in the complete population is:

[0137]

[0138] in, represents the kth hyperparameter encoding in the complete population, Indicates updated k=1,2,…,K, K represents the total number of hyperparameter encodings in the complete population, π represents pi, t represents the current optimization number, T represents the maximum optimization number, and r5 represents a random number between (0,1).

[0139] Although the previous search strategies have a certain ability to jump out of the local area, they are more likely to use individual information interaction to quickly explore unexplored areas, thereby helping to find the best position. In order to further improve the global search capability, an embodiment of the present invention provides a disturbance update that enables the algorithm to have the ability to jump out of the local optimum at any time.

[0140] Optionally, after each update of the hyperparameter encoding, it should also be processed for out-of-bounds errors to ensure that the hyperparameters are always valid.

[0141] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for monitoring electricity theft in charging piles based on big data, characterized in that: include: Obtaining historical electricity consumption big data corresponding to the charging pile and manual labels corresponding to the historical electricity consumption data, and cleaning the historical electricity consumption big data and manual labels corresponding to the historical electricity consumption data to obtain the cleaned historical electricity consumption big data and manual labels corresponding to the historical electricity consumption data; A deep learning model is used to build a charging pile electricity theft detection model, and the charging pile electricity theft detection model is optimized using the cleaned historical electricity consumption big data and the manual labels corresponding to the historical electricity consumption data to obtain the optimized charging pile electricity theft detection model; The optimized charging pile electricity theft detection model is deployed on the electricity consumption data monitoring platform, and the real-time electricity consumption data is analyzed through the deployed charging pile electricity theft detection model to determine the monitoring results of the charging pile electricity theft behavior; The historical electricity consumption big data corresponding to the charging pile includes electricity consumption type data and charging pile electric energy meter information; The artificial labels corresponding to the historical electricity consumption data include whether there is electricity theft at the charging pile or there is no electricity theft at the charging pile; Among them, the electricity consumption type data includes the time series data of daily electricity consumption of charging piles and the time series data of daily line loss in the substation area; Charging pile electric energy meter information includes electric energy meter current imbalance event record, electric energy meter overcurrent event record, electric energy meter constant magnetic field interference event record, electric energy meter cover opening event record, electric energy meter undervoltage event record, electric energy meter loss of current event record, electric energy meter loss of pressure event record and electric energy meter reverse event record; The charging pile electricity theft behavior detection model is optimized by using the cleaned historical electricity consumption big data and the manual labels corresponding to the historical electricity consumption data, and the optimized charging pile electricity theft behavior detection model is obtained, including: The chaotic mapping initialization method is used to initialize the hyperparameters of the charging pile electricity theft detection model multiple times to obtain multiple different hyperparameter encodings. According to the cleaned historical electricity consumption big data and the manual labels corresponding to the historical electricity consumption data, the error function value of each hyperparameter encoding is obtained; Determine the optimal hyperparameter encoding according to the error function value of each hyperparameter encoding; All hyperparameter encodings are randomly divided into two sub-populations; The first information interaction strategy is adopted to make the hyperparameter encodings in the two sub-populations perform initial information interaction to obtain the sub-populations after initial update; According to the optimal hyperparameter encoding, a second information interaction strategy is adopted to make the hyperparameter encodings in the two sub-populations perform secondary information interaction to obtain a sub-population after secondary update; According to the optimal hyperparameter encoding, the third information interaction strategy is adopted to make the hyperparameter encodings in the two sub-populations interact with each other three times, and the sub-populations are obtained after three updates; The two sub-populations after three updates are merged into a complete population, and the perturbation update is performed on each hyperparameter encoding in the complete population to obtain the updated complete population; Based on the updated complete population, the optimal hyperparameter encoding is re-obtained, and it is determined whether the current optimization times have reached the maximum optimization times. If so, the optimal hyperparameter encoding is used as the final hyperparameter of the charging pile electricity theft behavior detection model to obtain the optimized charging pile electricity theft behavior detection model. Otherwise, the step of obtaining the error function value of each hyperparameter encoding is returned; According to the cleaned historical electricity consumption big data and the manual labels corresponding to the historical electricity consumption data, the error function value encoded by each hyperparameter is obtained, including: After applying the hyperparameters included in the hyperparameter encoding to the charging pile electricity theft detection model, the cleaned historical electricity consumption big data is used as the actual input of the charging pile electricity theft detection model, and the manual labels corresponding to the historical electricity consumption data are used as the expected output of the charging pile electricity theft detection model. The mean square error or root mean square error corresponding to the charging pile electricity theft detection model is obtained, and the error function value of the hyperparameter encoding is obtained; Traverse all hyperparameter encodings and obtain the error function value of each hyperparameter encoding; The first information interaction strategy is used to make the hyperparameter encodings in the two sub-populations perform initial information interaction, and obtain the sub-population after the initial update, including: Pair the hyperparameter encodings in the two subpopulations one by one; For any pair of hyperparameter encodings, determine the hyperparameter encoding with the larger error function value And the hyperparameter encoding with a small error function value ; Hyperparameter encoding Encoding hyperparameters Paired individuals; Hyperparameter encoding for small error function values , directly encode it as the updated hyperparameter; Hyperparameter encoding for large error function values , which is encoded with the hyperparameter Perform information exchange to obtain updated hyperparameter encoding for: in, Represents hyperparameter encoding In the t +1 for the amount of updates during the optimization process, Represents the updated hyperparameter encoding ; Represents hyperparameter encoding In the t The update amount during the optimization process, when the t When there is no update amount during the optimization process, is 0; represents the inertia weight, represents a random number uniformly distributed in [0,1], represents a random number uniformly distributed in [0,1], represents the interaction control factor, Represents hyperparameter encoding The average code corresponding to the subpopulation; Traverse each pair of hyperparameter encodings to obtain the sub-population after the initial update; According to the optimal hyperparameter encoding, the second information interaction strategy is adopted to make the hyperparameter encodings in the two subpopulations perform secondary information interaction, and obtain the subpopulation after secondary update, including: Pair the hyperparameter encodings in the two subpopulations one by one; For any pair of hyperparameter encodings, determine the hyperparameter encoding with the larger error function value And the hyperparameter encoding with a small error function value ; Hyperparameter encoding Encoding hyperparameters Paired individuals; Hyperparameter encoding for small error function values , directly encode it as the updated hyperparameter; Hyperparameter encoding for large error function values , which is encoded with the hyperparameter Perform information exchange to obtain updated hyperparameter encoding for: in, Represents hyperparameter encoding In the t +1 for the amount of updates during the optimization process, Represents the updated hyperparameter encoding ; Represents hyperparameter encoding In the t The update amount during the optimization process, when the t When there is no update amount during the optimization process, is 0; represents the inertia weight, represents a random number uniformly distributed in [0,1], represents a random number uniformly distributed in [0,1], represents the interaction control factor, Represents hyperparameter encoding The average code corresponding to the subpopulation, represents a random number uniformly distributed in [0,1], represents the optimal hyperparameter encoding; Traverse each pair of hyperparameter encodings to obtain the sub-population after the second update; According to the optimal hyperparameter encoding, the third information interaction strategy is adopted to make the hyperparameter encodings in the two subpopulations interact with each other three times, and the subpopulations after three updates are obtained, including: Determine the current optimization times t, and determine the decision factor as t / T according to the current optimization times t; where T represents the maximum optimization times; For any pair of hyperparameter encodings, when the decision factor is less than or equal to 0.5, the hyperparameter encoding is updated as: in, and Paired hyperparameter encodings, Represents the updated hyperparameter encoding , Represents the updated hyperparameter encoding , Represents a random number between (0,1), represents the optimal hyperparameter encoding, represents the information interaction degree factor; For any pair of hyperparameter encodings, when the decision factor is greater than 0.5, the hyperparameter encoding is updated as: Traverse each pair of hyperparameter encodings and obtain three updated sub-populations; The two sub-populations after three updates are merged into a complete population, and the perturbation update is performed on each hyperparameter encoding in the complete population to obtain the updated complete population, including: Merge the two sub-populations after three updates into a complete population; The perturbation update for the hyperparameter encoding in the complete population is: in, Indicates the first k Hyperparameter encoding, Indicates updated , k=1,2,…,K, K represents the total number of hyperparameter encodings in the complete population, represents pi, Indicates the current number of optimizations. represents the maximum number of optimizations, Represents a random number between (0,1).

2. The method for monitoring electricity theft behavior of charging piles based on big data according to claim 1 is characterized in that: Clean the historical electricity consumption big data and the manual labels corresponding to the historical electricity consumption data, and obtain the cleaned historical electricity consumption big data and the manual labels corresponding to the historical electricity consumption data, including: The historical electricity consumption big data with abnormal values ​​and / or missing values ​​and the artificial labels corresponding to the historical electricity consumption data are deleted to obtain the cleaned historical electricity consumption big data and the artificial labels corresponding to the historical electricity consumption data.

3. The method for monitoring electricity theft in charging piles based on big data according to claim 1 is characterized in that: A deep learning model is used to build a charging pile electricity theft behavior detection model, including: using a convolutional neural network to build a charging pile electricity theft behavior detection model.

Images