Artificial intelligence-based energy storage system energy management method and device

Abstract

The application discloses an energy management method and device for an energy storage system based on artificial intelligence, relates to the technical field of energy management, and constructs samples by collecting multidimensional historical power consumption correlation data, improves the accuracy of power consumption prediction, makes energy management of the energy storage system more accurate, dynamically adjusts charging power according to the predicted power consumption and the residual power of the energy storage system, not only realizes fine management of energy of the energy storage system, improves energy utilization efficiency, but also prolongs the service life of the energy storage system, finally charges at a low electricity price valley period, reduces the operation cost of the energy storage system, and improves economic benefits.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy management, in particular to an energy management method and device for energy storage system based on artificial intelligence. BACKGROUND

[0002] With the transformation of energy structure and large-scale application of renewable energy, the role of energy storage system in power system is increasingly prominent. Energy storage system can store energy when power demand is low and release energy when power demand is high, thereby balancing the load of power grid and improving energy utilization efficiency. However, the traditional energy management method of energy storage system mainly relies on fixed scheduling strategy or simple prediction model, which is difficult to adapt to complex and changeable power consumption environment, resulting in low energy management efficiency, and the peak load shifting effect of energy storage system cannot be fully played. Therefore, it is difficult to balance the service life and efficiency. SUMMARY

[0003] The present application provides an energy management method and device for energy storage system based on artificial intelligence, aiming to solve the problem that the prior art cannot balance the energy management efficiency and service life.

[0004] The first aspect of the present application provides an energy management method for energy storage system based on artificial intelligence, comprising:

[0005] Collecting historical power consumption correlation data corresponding to the energy supply area corresponding to the energy storage system, and constructing historical load influence samples and their corresponding power consumption labels according to the historical power consumption correlation data;

[0006] Using an artificial intelligence model to learn the data relationship between the historical load influence samples and their corresponding power consumption labels, and obtaining a power consumption prediction model;

[0007] Collecting real-time load influence samples of the energy supply area, and scheduling the power consumption prediction model to identify the real-time load influence samples to determine the predicted power consumption of the energy supply area;

[0008] Determining the remaining power of the energy storage system, and determining the charging power corresponding to the energy storage system according to the remaining power and the predicted power consumption;

[0009] Based on the charging power corresponding to the energy storage system, the energy management of the energy storage system is carried out, and the energy management of the energy storage system based on artificial intelligence is completed.

[0010] In a possible implementation, the historical power consumption correlation data corresponding to the energy supply area corresponding to the energy storage system is collected, and a historical load influence sample and a power consumption label corresponding thereto are constructed according to the historical power consumption correlation data, including:

[0011] On the basis of a preset data sampling frequency, historical power consumption, the highest temperature of the day, the lowest temperature of the day, the average temperature of the day, the season of the day, date information, holiday information, and weather information of the day corresponding to the energy supply area corresponding to the energy storage system are collected at a plurality of sampling time points to obtain historical power consumption correlation data;

[0012] The historical load influence sample is constructed by collecting the historical power consumption correlation data at N continuous sampling time points, and the historical power consumption at the N+1 sampling time point is taken as the power consumption label.

[0013] In a possible implementation, a data relationship between the historical load influence sample and the power consumption label corresponding thereto is learned by using an artificial intelligence model to obtain a power consumption prediction model, including:

[0014] Random initialization coding is performed between the upper and lower limits of the model parameters of the artificial intelligence model to obtain model parameter coding, and a plurality of different model parameter codings are repeatedly obtained;

[0015] On the basis of the historical load influence sample and the power consumption label corresponding thereto, the fitness corresponding to all model parameter codings is obtained;

[0016] According to the fitness corresponding to the model parameter coding, a multi-parameter adaptive local search mechanism is used to adaptively search the model parameter coding to obtain the model parameter coding after adaptive search;

[0017] An adaptive collaborative search mechanism is used to collaboratively search the model parameter coding after adaptive search to obtain the model parameter coding after collaborative search;

[0018] An adaptive number of mutation fusion mechanisms are used to search the model parameter coding after collaborative search to obtain the model parameter coding after mutation fusion search;

[0019] The adaptive search, the collaborative search, and the mutation fusion search are repeated until the maximum training number is reached, the model parameter coding after mutation fusion search is obtained, and the target optimal parameter coding is obtained;

[0020] The power consumption prediction model is obtained according to the target optimal parameter coding.

[0021] In one possible implementation, based on the fitness corresponding to the model parameter encoding, a multi-parameter adaptive local search mechanism is used to adaptively search the model parameter encoding to obtain the adaptively searched model parameter encoding, including:

[0022] Based on the fitness corresponding to the model parameter encoding and the current number of training iterations, the adaptive speed adjustment factor is determined as follows:

[0023]

[0024] in, Indicates the first k Each model parameter is encoded with an adaptive speed adjustment factor. This represents the maximum value of the adaptive speed adjustment factor. Indicates the maximum number of training iterations. Indicates the current number of training iterations. This represents the fitness corresponding to the encoding of the k-th model parameter. This represents the maximum fitness among all encoded model parameters;

[0025] Based on the fitness corresponding to the model parameter encoding and the current number of training iterations, the first local learning factor and the second local learning factor are determined as follows:

[0026]

[0027] in, Represents the first local learning factor. This represents the second local learning factor. Represents the cosine function. Represents pi;

[0028] Based on the adaptive speed adjustment factor, the first local learning factor, and the second local learning factor, obtain the first... k The search speed corresponding to the encoding of each model parameter is:

[0029]

[0030] in, Indicates the first t During the training process, the first k The search speed corresponding to each model parameter encoding Indicates the first t +1 training session k The search speed corresponding to each model parameter encoding k =1,2,...,K, K This indicates the total number of model parameter codes. Represents the first random number between (0,1). a second random number between (0, 1), a first model parameter encoding corresponding to a historical optimal value, k a second model parameter encoding corresponding to a historical optimal value, a first model parameter encoding, k a second model parameter encoding, a model parameter encoding corresponding to a maximum fitness value;

[0031] performing adaptive search on the model parameter encodings according to the search speed, to obtain model parameter encodings after adaptive search as:

[0032]

[0033] wherein, a first model parameter encoding after adaptive search, k a second model parameter encoding after adaptive search.

[0034] In a possible implementation, a cooperative search mechanism is adopted to perform cooperative search on the model parameter encodings after adaptive search, to obtain model parameter encodings after cooperative search, including:

[0035] for model parameters of any one dimension, average the model parameters of all the model parameter encodings after adaptive search, and construct the average values corresponding to all dimensions into a vector to obtain a center encoding;

[0036] arrange the model parameter encodings after adaptive search in an order from small to large fitness value, to obtain model parameter encodings after arrangement;

[0037] for the model parameter encodings after arrangement, the first cooperative search factor and the second cooperative search factor are:

[0038]

[0039]

[0040] wherein, a first cooperative search factor corresponding to a first model parameter encoding after arrangement, i a first cooperative search factor corresponding to a first model parameter encoding after arrangement, a second cooperative search factor corresponding to a second model parameter encoding after arrangement, i a second cooperative search factor corresponding to a second model parameter encoding after arrangement, sin a sine function, cos a cosine function, a cooperative search direction control parameter, and , a constant term, a third random number between (0, 1), a cooperative search range control parameter, and , denotes a search range base influence parameter, denotes a fourth random number between (0, 1); denotes all

[0041] According to the current training number, the position disturbance factor is obtained as:

[0042]

[0043] wherein, denotes a position disturbance factor, denotes a circular constant, denotes a maximum value of the position disturbance factor, denotes a minimum value of the position disturbance factor, denotes a maximum training number;

[0044] According to the center code, the position disturbance factor, the first cooperative search factor and the second cooperative search factor, the model parameter code after permutation is cooperatively searched to obtain the model parameter code after cooperative search as:

[0045]

[0046] wherein, denotes the model parameter code after the i th permutation, i =1, 2,..., K, K denotes a total number of model parameter codes, denotes the model parameter code after the i +1th permutation, and i when =K, then is the model parameter code after random permutation, denotes the model parameter code after the i th cooperative search, denotes a center code.

[0047] In a possible implementation, the model parameter code after cooperative search is searched by using an adaptive number of mutation fusion mechanisms to obtain the model parameter code after mutation fusion search, comprising:

[0048] According to the fitness of the model parameter code after cooperative search, the adaptive mutation fusion number is obtained as:

[0049]

[0050] wherein, denotes the model parameter code after the mThe number of adaptive mutation fusions corresponding to the model parameter encoding after each collaborative search. m =1,2,...,K, K This indicates the total number of model parameter codes. This represents the floor function. Indicates the first m The fitness of the model parameters after a collaborative search is encoded. This represents the maximum fitness among all encoded model parameters. This represents the minimum fitness among all the encoded model parameters;

[0051] For the first m The model parameter encoding after each collaborative search is randomly selected. The model parameters are encoded after another collaborative search, resulting in... One variant fusion code;

[0052] According to the above The model parameter codes after the cooperative search are subjected to mutation fusion search to obtain the model parameter codes after mutation fusion search as follows:

[0053]

[0054] in, Indicates the first m Model parameter encoding after a collaborative search Indicates the first m Model parameter encoding after mutation fusion search, This represents the fifth random number between (0,1). Represents model parameter encoding The corresponding number j One variant fusion code, This represents the encoding of the model parameters corresponding to the maximum fitness.

[0055] In one possible implementation, real-time load impact samples of the energy supply area are collected, and the electricity consumption prediction model is scheduled to identify the real-time load impact samples to determine the predicted electricity consumption of the energy supply area, including:

[0056] Collect real-time load impact samples of the energy supply area; wherein, the data structure of the real-time load impact samples is the same as that of the historical load impact samples;

[0057] The real-time load impact sample is used as the input to the electricity consumption prediction model to obtain the predicted electricity consumption of the energy supply area.

[0058] In a possible implementation, the remaining power of the energy storage system is determined, and the corresponding charging power of the energy storage system is determined according to the remaining power and the predicted power consumption.

[0059] The remaining power of the energy storage system is determined.

[0060] The power difference is obtained according to the difference between the remaining power and the predicted power consumption.

[0061] In a case where the power difference is greater than or equal to zero, the corresponding charging power of the energy storage system is determined as a first preset charging power.

[0062] In a case where the power difference is less than zero, a second preset charging power corresponding to the power difference is determined, wherein the second preset charging power is greater than the first preset charging power; different power differences are preset with different second preset charging powers, and the power difference is directly proportional to the second preset charging power, and the first preset charging power and the second preset charging power are both less than or equal to the rated charging power designed for the energy storage system.

[0063] In a possible implementation, the energy storage system is managed based on the corresponding charging power of the energy storage system, including:

[0064] The energy storage system is controlled to charge in a preset time period based on the corresponding charging power of the energy storage system, wherein the preset time period refers to a low-price time period.

[0065] The second aspect of the present application provides an energy storage system energy management device based on artificial intelligence, including a data acquisition module, a data training module, a power consumption prediction module, a charging determination module, and an energy management module.

[0066] The data acquisition module is configured to acquire historical power consumption correlation data corresponding to an energy supply area corresponding to an energy storage system, and construct historical load influence samples and corresponding power consumption labels according to the historical power consumption correlation data.

[0067] The data training module is configured to learn the data relationship between the historical load influence samples and the corresponding power consumption labels by using an artificial intelligence model, to obtain a power consumption prediction model.

[0068] The power consumption prediction module is configured to acquire real-time load influence samples of the energy supply area, and schedule the power consumption prediction model to identify the real-time load influence samples, to determine the predicted power consumption of the energy supply area.

[0069] The charging determination module is configured to determine the remaining power of the energy storage system, and determine the charging power corresponding to the energy storage system according to the remaining power and the predicted power consumption.

[0070] The energy management module is configured to perform energy management on the energy storage system based on the charging power corresponding to the energy storage system, and complete the energy management of the energy storage system based on artificial intelligence.

[0071] Advantages:

[0072] The present application provides an energy management method and device for an energy storage system based on artificial intelligence. By collecting multi-dimensional historical power consumption correlation data to construct samples, the accuracy of power consumption prediction is improved, the energy management of the energy storage system is more accurate, and the charging power is dynamically adjusted according to the predicted power consumption and the remaining power of the energy storage system. Not only the fine management of the energy of the energy storage system is realized, the energy utilization efficiency is improved, but also the service life of the energy storage system is prolonged. Finally, the energy storage system is charged at the low price valley period, the operation cost of the energy storage system is reduced, and the economic benefit is improved. BRIEF DESCRIPTION OF DRAWINGS

[0073] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the description of the embodiments of the present application. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0074] Figure 1 is a flow chart of an energy management method for an energy storage system based on artificial intelligence according to an embodiment of the present application;

[0075] Figure 2 is a structural diagram of an energy management device for an energy storage system based on artificial intelligence according to an embodiment of the present application;

[0076] Legend: 201-data acquisition module, 202-data training module, 203-power consumption prediction module, 204-charging determination module, 205-energy management module. DETAILED DESCRIPTION

[0077] The technical solutions of the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0078] As Figure 1As shown, the embodiment of the present application provides an energy management method for a storage system based on artificial intelligence, which comprises the following steps:

[0079] S101, historical power consumption correlation data corresponding to a power supply area corresponding to a storage system is collected, and a historical load influence sample and its corresponding power consumption label are constructed according to the historical power consumption correlation data.

[0080] For example, the historical power consumption correlation data includes but is not limited to historical power consumption, temperature data, seasonal information, date information, holiday information and weather information, etc. These data reflect the influencing factors of power consumption from different angles, and through comprehensive analysis of these data, the future power consumption can be more accurately predicted.

[0081] S102, an artificial intelligence model is used to learn the data relationship between the historical load influence sample and its corresponding power consumption label, and a power consumption prediction model is obtained.

[0082] The historical load influence sample usually has an enlightening effect on power consumption. By learning the data relationship between the historical load influence sample and its corresponding power consumption label through an artificial intelligence model, the artificial intelligence model can have the ability to predict power consumption, thereby obtaining a power consumption prediction model.

[0083] S103, real-time load influence samples of the power supply area are collected, and the power consumption prediction model is used to identify the real-time load influence samples to determine the predicted power consumption of the power supply area.

[0084] The data structure of the real-time load influence sample of the power supply area should be the same as that of the historical load influence sample, so as to ensure that the power consumption prediction model can accurately identify the real-time load influence sample and obtain the predicted power consumption of the power supply area.

[0085] S104, the remaining power of the storage system is determined, and the charging power corresponding to the storage system is determined according to the remaining power and the predicted power consumption.

[0086] The greater the difference between the remaining power and the predicted power consumption, the greater the required charging power, thereby ensuring that the peak load shifting and valley filling effect can be achieved. The smaller the difference between the remaining power and the predicted power consumption, the smaller the required charging power, thereby ensuring the service life of the storage system and avoiding the reduction of service life caused by long-time high-power charging.

[0087] S105, based on the charging power corresponding to the storage system, the energy management of the storage system is performed to complete the energy management of the storage system based on artificial intelligence.

[0088] Based on the charging power corresponding to the energy storage system, the energy storage system can be supplemented with energy during the low electricity price period. The low electricity price period generally refers to 23:00 in the evening to 07:00 in the morning of the next day. Charging during this period can effectively reduce the cost of energy supplement.

[0089] The present application provides an energy storage system energy management method based on artificial intelligence. By collecting multi-dimensional historical electricity consumption correlation data to construct samples, the accuracy of electricity consumption prediction is improved, making the energy storage system energy management more accurate. According to the predicted electricity consumption and the remaining electricity of the energy storage system, the charging power is dynamically adjusted, not only realizing the fine management of the energy of the energy storage system, improving the energy utilization efficiency, but also prolonging the service life of the energy storage system. Finally, charging is performed during the low electricity price period, reducing the operation cost of the energy storage system and improving the economic benefit.

[0090] In one possible implementation, historical electricity consumption correlation data corresponding to the energy supply area corresponding to the energy storage system is collected, and historical load influence samples and their corresponding electricity consumption labels are constructed according to the historical electricity consumption correlation data, including:

[0091] Based on a preset data sampling frequency, historical electricity consumption, the highest temperature of the day, the lowest temperature of the day, the average temperature of the day, the day of the season (such as a unique code corresponding to spring, summer, autumn and winter), date information (such as year, month and day), holiday information (such as a unique code corresponding to holidays and non-holidays), and weather information of the day (a unique code corresponding to different weather, such as a unique code corresponding to light rain, and a unique code corresponding to shower rain) corresponding to the energy supply area corresponding to the energy storage system are collected at multiple sampling time points to obtain historical electricity consumption correlation data.

[0092] The historical load influence samples are constructed by collecting historical electricity consumption correlation data at N consecutive sampling time points, and the historical electricity consumption at the N+1 sampling time point is taken as the electricity consumption label.

[0093] The data sampling frequency can be set to 1 day collection. Then, the real-time load influence sample refers to collecting historical electricity consumption correlation data at N consecutive sampling time points closest to the prediction time point (usually in units of days) that needs to be predicted.

[0094] It should be noted that the data length should be kept consistent for any kind of data, so that the data size or data length of the historical load impact sample constructed is consistent. For example, assuming that the historical power consumption correlation data at any sampling time point is constructed as a row of data, the historical load impact sample is a data matrix. By unifying the different numerical lengths of one kind of data, it can be ensured that the size of all historical power consumption correlation data is the same. Assuming that there is a three-digit data and a four-digit data in the historical power consumption, the three-digit data can be padded with zeros in front to make the length of the historical power consumption data at different sampling time points the same. It should be noted that for non-digital data such as the day of the season, holiday information and weather information of the day, an encoding method can be used to represent it, and a unique code with consistent length is assigned to each type.

[0095] In recent years, artificial intelligence technology has been widely applied in energy management field. Through machine learning algorithm, historical power consumption data is analyzed and predicted, which can improve the accuracy of power consumption prediction. However, the existing artificial intelligence prediction model has problems such as slow convergence speed and easy to fall into local optimum in the training process, which affects the prediction accuracy. At the same time, in the energy management strategy of energy storage system, the existing technology often lacks dynamic adjustment mechanism and cannot perform fine control according to the predicted power consumption and the actual state of the energy storage system. Therefore, the embodiments of the present application improve the defects of the prior art, as follows.

[0096] In one possible implementation, an artificial intelligence model is used to learn the data relationship between the historical load impact sample and the corresponding power consumption label, and a power consumption prediction model is obtained, including:

[0097] Randomly initialize the upper and lower limits of the model parameters of the artificial intelligence model to obtain model parameter codes, and repeat to obtain multiple different model parameter codes.

[0098] The artificial intelligence model can be set as a convolutional neural network, and the model parameters of the convolutional neural network can be connection weights between different network layers. The model parameters can be randomly initialized between the upper and lower limits of the model parameters and coded into vectors to obtain model parameter codes.

[0099] Based on the historical load impact sample and the corresponding power consumption label, the fitness of all model parameter codes is obtained.

[0100] For example, the historical load impact sample in the historical load impact sample can be taken as input, and the power consumption label can be taken as expected output to obtain a cross-entropy loss function value. The cross-entropy loss function value is added to 0.0001 and takes the reciprocal to obtain the fitness of all model parameter codes.

[0101] Based on the fitness corresponding to the model parameter encoding, a multi-parameter adaptive local search mechanism is used to adaptively search the model parameter encoding to obtain the adaptively searched model parameter encoding.

[0102] An adaptive collaborative search mechanism is used to perform a collaborative search on the model parameter encoding after the adaptive search, resulting in the model parameter encoding after collaborative search.

[0103] An adaptive number of mutation fusion mechanisms are used to perform mutation fusion search on the model parameter encoding after the cooperative search, resulting in the model parameter encoding after mutation fusion search.

[0104] Repeat adaptive search, collaborative search, and mutation fusion search until the maximum number of training iterations is reached. Obtain the model parameter encoding after mutation fusion search and obtain the target optimal parameter encoding.

[0105] For example, during the Tth training process, the model parameter encoding with the highest fitness can be obtained as the target optimal parameter encoding based on the model parameter encoding after mutation fusion search.

[0106] Based on the target optimal parameter encoding, the electricity consumption prediction model is obtained, that is, the model parameters in the target optimal parameter encoding are used as the target model parameters of the artificial intelligence model to obtain the electricity consumption prediction model.

[0107] Optionally, after each processing of the model parameter encoding, an out-of-bounds handling process can be performed to ensure that the model parameter encoding remains within the valid range.

[0108] In one possible implementation, based on the fitness corresponding to the model parameter encoding, a multi-parameter adaptive local search mechanism is used to adaptively search the model parameter encoding to obtain the adaptively searched model parameter encoding, including:

[0109] Based on the fitness corresponding to the model parameter encoding and the current number of training iterations, the adaptive speed adjustment factor is determined as follows:

[0110]

[0111] in, Indicates the first k Each model parameter is encoded with an adaptive speed adjustment factor. This represents the maximum value of the adaptive speed adjustment factor. Indicates the maximum number of training iterations. Indicates the current number of training iterations. This represents the fitness corresponding to the encoding of the k-th model parameter. This represents the maximum fitness among all encoded model parameters;

[0112] According to the fitness corresponding to the model parameter code and the current training number, the first local learning factor and the second local learning factor are determined as follows:

[0113]

[0114] wherein, denotes the first local learning factor, denotes the second local learning factor, denotes a cosine function, denotes a constant of a circle;

[0115] According to the adaptive speed adjustment factor, the first local learning factor and the second local learning factor, the search speed corresponding to the first model parameter code is obtained as follows: k

[0116]

[0117] wherein, denotes the search speed corresponding to the first model parameter code in the first training process, t denotes the search speed corresponding to the first model parameter code in the first training process, k denotes the search speed corresponding to the first model parameter code in the first training process, denotes the search speed corresponding to the first model parameter code in the first training process, t =1, 2,..., K, k denotes the total number of model parameter codes, k denotes a first random number between (0, 1), K denotes a second random number between (0, 1), denotes the historical optimal value corresponding to the first model parameter code, denotes the first model parameter code, denotes the model parameter code with the maximum fitness. k According to the search speed, the model parameter code is adaptively searched to obtain the model parameter code after adaptive search as follows: k

[0118] wherein,

[0119] denotes the model parameter code after adaptive search.

[0120] k

[0121] ​​​​​​The adaptive local search mechanism is adopted to adaptively search the model parameter coding, and the adaptive search can be performed based on the fitness of the model parameter coding and the number of training, so that a larger range of search can be performed in the case of excessive aggregation, and otherwise a more fine search is performed, thereby ensuring the training diversity. With the algorithm, the information learning intensity of the model parameter coding with the historical optimal quality and the maximum fitness is gradually reduced, the search precision of the algorithm is improved, and the convergence of the algorithm is ensured.

[0122] In a possible implementation, an adaptive cooperative search mechanism is adopted to cooperatively search the model parameter coding after the adaptive search, to obtain the model parameter coding after the cooperative search, including:

[0123] For the model parameters of any dimension, the model parameters of all the model parameter coding after the adaptive search are averaged, and the average values corresponding to all dimensions are constructed as a vector to obtain a center coding;

[0124] The model parameter coding after the adaptive search is arranged in order from small to large fitness to obtain the model parameter coding after the arrangement;

[0125] For the model parameter coding after the arrangement, the first cooperative search factor and the second cooperative search factor are obtained as:

[0126]

[0127]

[0128] wherein, the first cooperative search factor corresponding to the i-th model parameter coding after the arrangement is represented by i the second cooperative search factor corresponding to the i-th model parameter coding after the arrangement is represented by i sin the sine function is represented by cos the cosine function is represented by the cooperative search direction control parameter is represented by , the constant term is represented by 6; the third random number between 0 and 1 is represented by the cooperative search range control parameter is represented by , the search range base influence parameter is represented by 1.5; the fourth random number between 0 and 1 is represented by the maximum value is determined in all the corresponding to the model parameter coding after the arrangement,​​ This represents the encoding of the model parameters after all permutations. The minimum value is determined in the middle.

[0129] Based on the current number of training iterations, the position perturbation factor is obtained as follows:

[0130]

[0131] in, Indicates the location disturbance factor. Represents pi (π). This represents the maximum value of the position disturbance factor, which can be set to 0.91; This represents the minimum value of the position disturbance factor, which can be set to 0.4; Indicates the maximum number of training iterations;

[0132] Based on the center code, position perturbation factor, first cooperative search factor, and second cooperative search factor, a cooperative search is performed on the arranged model parameter codes to obtain the cooperatively searched model parameter codes as follows:

[0133]

[0134] in, Indicates the first i The model parameter encoding after permutation, i =1,2,...,K, K This indicates the total number of model parameter codes. Indicates the first i +1 permutation of model parameter encoding, and i When =K, then Encode the model parameters after random permutation. Indicates the first i Model parameter encoding after a collaborative search Indicates the center code.

[0135] The embodiments of this application employ an adaptive collaborative search mechanism to perform collaborative search on the model parameter encoding after the adaptive search, which enables efficient collaborative search of the model parameter encoding, thereby exploring more solution space and facilitating the discovery of better solutions.

[0136] In one possible implementation, an adaptive number of mutation fusion mechanisms is used to perform mutation fusion search on the model parameter encoding after the cooperative search, resulting in the model parameter encoding after mutation fusion search, including:

[0137] Based on the fitness of the model parameters encoded after collaborative search, the number of adaptive mutation fusions is obtained as follows:

[0138]

[0139] wherein, denotes the model parameter encoding after the kth collaborative search, m denotes the adaptive number of fusion of the model parameter encoding corresponding to the kth collaborative search, m =1,2,...,K, K denotes the total number of model parameter encodings, denotes the ceiling function, denotes the model parameter encoding after the kth collaborative search, m denotes the fitness of the model parameter encoding corresponding to the kth collaborative search, denotes the maximum fitness of all model parameter encodings, denotes the minimum fitness of all model parameter encodings.

[0140] randomly selects, for the model parameter encoding after the kth collaborative search, m model parameter encodings after other collaborative searches, to obtain K fusion encodings; According to the K fusion encodings, the model parameter encoding after the collaborative search is subjected to fusion search to obtain a model parameter encoding after fusion search:

[0141]

[0142]

[0143] wherein, denotes the model parameter encoding after the kth collaborative search, m denotes the model parameter encoding after the kth fusion search, denotes the fifth random number between (0, 1), m denotes the model parameter encoding corresponding to the kth fusion encoding, denotes the model parameter encoding corresponding to the maximum fitness. j

[0144] The embodiments of the present application adopt the adaptive number of fusion mechanism to perform fusion search on the model parameter encoding after the collaborative search, thereby providing the algorithm with strong global search capability, enabling the algorithm to escape from the current optimal position and search for other positions in the solution space, so as to avoid the algorithm from falling into local optimum. The fusion search can also be controlled by using the greedy strategy to ensure the training speed of the algorithm.

[0145] Through the mutual cooperation of the above-mentioned mechanisms, the power consumption prediction model obtained by training can have more accurate prediction capability, thereby ensuring the accuracy of energy management.​​​​​​​

[0146] In a possible implementation, a real-time load influence sample of the energy supply area is collected, and the power consumption prediction model is scheduled to identify the real-time load influence sample to determine the predicted power consumption of the energy supply area, including:

[0147] The real-time load influence sample of the energy supply area is collected. The data structure of the real-time load influence sample is the same as that of the historical load influence sample.

[0148] The real-time load influence sample is taken as an input of the power consumption prediction model to obtain the predicted power consumption of the energy supply area.

[0149] Optionally, the real-time load influence sample and the historical load influence sample can be normalized and then taken as the input of the power consumption prediction model.

[0150] In a possible implementation, the residual power of the energy storage system is determined, and the corresponding charging power of the energy storage system is determined according to the residual power and the predicted power consumption, including:

[0151] The residual power of the energy storage system is determined.

[0152] The difference between the residual power and the predicted power consumption is obtained as a power difference.

[0153] In a case where the power difference is greater than or equal to zero, the corresponding charging power of the energy storage system is determined as a first preset charging power.

[0154] In a case where the power difference is less than zero, a second preset charging power corresponding to the power difference is determined. The second preset charging power is greater than the first preset charging power. Different power differences are preset with different second preset charging powers, and the power difference is directly proportional to the second preset charging power. The first preset charging power and the second preset charging power are both less than or equal to the rated charging power designed for the energy storage system.

[0155] For example, the first preset charging power can be the power corresponding to the trickle charging phase. In a case where the residual power is sufficient, small-power charging can be used to avoid the life loss of the energy storage system.

[0156] Optionally, the second preset charging power maximum value should be set as the rated charging power designed for the energy storage system, that is, if the remaining power is less than the predicted power consumption, it indicates that the energy storage system needs to be powered to realize the full power supply of the power supply area. For example, if the predicted power consumption is greater than the total power of the energy storage system after being fully charged, the rated charging power can be directly used for charging to ensure that the energy storage system can charge as much as possible during the low price period. In the case that the predicted power consumption is less than or equal to the total power of the energy storage system, the second preset charging power can be set as the minimum power that meets the power consumption condition; the power consumption condition is set as the first sum value between the charging amount during the low price period and the remaining power being greater than or equal to the second sum value between the predicted power consumption and the preset fault tolerance power, and the preset fault tolerance power is the power fluctuation value set by the staff in advance, so that the actual power consumption can fluctuate upwards within a certain range and meet the power supply demand. If the first sum value cannot be greater than or equal to the second sum value when the second preset charging power is set as the rated charging power, the rated charging power is directly started to charge, so as to reduce the power purchased from the power grid during the non-low price period as much as possible, thereby improving the economy.

[0157] In a possible implementation, the energy management of the energy storage system is based on the charging power corresponding to the energy storage system, and includes:

[0158] The energy management of the energy storage system is based on the charging power corresponding to the energy storage system, and includes:

[0159] As shown in Figure 2 Based on the same inventive concept, the embodiments of the present application provide an energy storage system energy management device based on artificial intelligence, which includes a data acquisition module 201, a data training module 202, a power consumption prediction module 203, a charging determination module 204, and an energy management module 205.

[0160] The data acquisition module 201 is configured to acquire historical power consumption correlation data corresponding to a power supply area corresponding to an energy storage system, and construct a historical load influence sample and a power consumption label corresponding thereto according to the historical power consumption correlation data.

[0161] The data training module 202 is configured to learn the data relationship between the historical load influence sample and the power consumption label corresponding thereto by using an artificial intelligence model, and obtain a power consumption prediction model.

[0162] The power consumption prediction module 203 is configured to acquire a real-time load influence sample of the power supply area, and schedule the power consumption prediction model to identify the real-time load influence sample, and determine the predicted power consumption of the power supply area.

[0163] The charging determination module 204 is configured to determine the residual power of the energy storage system, and determine the charging power corresponding to the energy storage system according to the residual power and the predicted power consumption.

[0164] The energy management module 205 is configured to perform energy management on the energy storage system based on the charging power corresponding to the energy storage system, to complete the energy management of the energy storage system based on artificial intelligence.

[0165] The energy management device of the energy storage system based on artificial intelligence provided by the embodiments of the present application can execute the above-mentioned method technical solutions, and has similar principles and beneficial effects, which will not be described here.

[0166] Each embodiment in the specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same and similar parts between the embodiments can be referred to each other.

[0167] The embodiments of the present application are described with reference to the flowcharts and / or block diagrams of the methods, devices, electronic equipment and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing terminal equipment to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal equipment realize the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one flow or multiple flows and / or blocks Figure 1 The devices for realizing the functions specified in one block or multiple blocks.

[0168] These computer program instructions can also be stored in a computer readable memory which can guide the computer or other programmable data processing terminal equipment to work in a specific way, so that the instructions stored in the computer readable memory produce a product including instruction devices, which realize the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one flow or multiple flows and / or blocks Figure 1 The functions specified in one block or multiple blocks.

[0169] These computer program instructions can also be loaded into the computer or other programmable data processing terminal equipment, so that a series of operation steps are executed on the computer or other programmable terminal equipment to produce a computer implemented process, so that the instructions executed on the computer or other programmable terminal equipment provide the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one flow or multiple flows and / or blocks Figure 1 The functions specified in one block or multiple blocks.

[0170] While preferred embodiments of the application have been described, those skilled in the art will appreciate that other modifications than those specifically described can be made within the scope of the application. Accordingly, all changes coming within the meaning and equivalency range of the appended claims are intended to be embraced therein.

[0171] Finally, it should be noted that, in the present document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily implying any actual relationship or order between such entities or actions. Also, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0172] The principles and implementations of the present application have been described above with the specific examples. The above description of the embodiments is only for the purpose of understanding the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation and application range can be changed. In summary, the content of the present description should not be understood as a limitation of the present application.

Claims

1. An artificial intelligence-based energy management method for an energy storage system, characterized by, The method comprises the following steps: Collecting historical power consumption correlation data corresponding to the energy supply area corresponding to the energy storage system, and constructing historical load influence samples and their corresponding power consumption labels according to the historical power consumption correlation data; Learning the data relationship between the historical load influence samples and their corresponding power consumption labels by using an artificial intelligence model to obtain a power consumption prediction model; Collecting real-time load influence samples of the energy supply area, and scheduling the power consumption prediction model to identify the real-time load influence samples to determine the predicted power consumption of the energy supply area; Determining the remaining power of the energy storage system, and determining the charging power corresponding to the energy storage system according to the remaining power and the predicted power consumption; Based on the charging power corresponding to the energy storage system, the energy management of the energy storage system is completed, and the energy management of the energy storage system based on artificial intelligence is completed; Learning the data relationship between the historical load influence samples and their corresponding power consumption labels by using an artificial intelligence model to obtain a power consumption prediction model, comprising: Randomly initializing the coding between the upper and lower limits of the model parameters of the artificial intelligence model to obtain model parameter codes, and repeatedly obtaining multiple different model parameter codes; Based on the historical load influence samples and their corresponding power consumption labels, the fitness of all model parameter codes is obtained; According to the fitness of the model parameter codes, a multi-parameter adaptive local search mechanism is used to adaptively search the model parameter codes to obtain the model parameter codes after adaptive search; An adaptive collaborative search mechanism is used to collaboratively search the model parameter codes after adaptive search to obtain the model parameter codes after collaborative search; An adaptive number of mutation fusion mechanisms are used to perform mutation fusion search on the model parameter codes after collaborative search to obtain the model parameter codes after mutation fusion search; Repeat the adaptive search, collaborative search and mutation fusion search until the maximum training number is reached, obtain the model parameter codes after mutation fusion search, and obtain the target optimal parameter code; According to the target optimal parameter code, a power consumption prediction model is obtained.

2. The artificial intelligence-based energy storage system energy management method of claim 1, wherein, Collecting historical power consumption correlation data corresponding to the energy supply area corresponding to the energy storage system, and constructing historical load influence samples and their corresponding power consumption labels according to the historical power consumption correlation data, comprising: Based on a preset data sampling frequency, historical power consumption, daily maximum temperature, daily minimum temperature, daily average temperature, daily season, date information, holiday information and daily weather information of the energy supply area corresponding to the energy storage system are collected at multiple sampling time points to obtain historical power consumption correlation data; Collecting historical power consumption correlation data at N consecutive sampling time points to construct historical load influence samples, and taking the historical power consumption at the N+1 sampling time point as the power consumption label.

3. The artificial intelligence-based energy storage system energy management method of claim 1, wherein, According to the fitness of the model parameter codes, a multi-parameter adaptive local search mechanism is used to adaptively search the model parameter codes to obtain the model parameter codes after adaptive search, comprising: According to the fitness of the model parameter codes and the current training number, the adaptive speed adjustment factor is determined as: wherein, represents the kth model parameter encoding corresponding adaptive speed adjustment factor, k represents the kth model parameter encoding corresponding adaptive speed adjustment factor, represents the maximum value of adaptive speed adjustment factor, represents the maximum training number, represents the current training number, represents the kth model parameter encoding corresponding fitness, represents the maximum fitness in all model parameter encodings; According to the model parameter code corresponding to the fitness and the current training number, the first local learning factor and the second local learning factor are determined as: wherein denotes a first local learning factor, denotes a second local learning factor, denotes a cosine function, denotes the circle constant; According to the adaptive speed adjustment factor, the first local learning factor, and the second local learning factor, a first model parameter encoding corresponding search speed is obtained as follows: k ​ wherein, denotes the t th model parameter encoding corresponding search speed in the k th training process, denotes the t th model parameter encoding corresponding search speed in the k th training process, k = 1, 2,..., K, K denotes the total number of model parameter encodings, denotes a first random number between (0, 1), denotes a second random number between (0, 1), denotes the k th model parameter encoding corresponding historical optimal value, denotes the k th model parameter encoding, denotes the model parameter encoding corresponding to the maximum fitness. According to the search speed, the model parameter code is adaptively searched to obtain the model parameter code after adaptive search as: wherein, represents the model parameter encoding after the k adaptive search.

4. The artificial intelligence-based energy storage system energy management method of claim 3, wherein, An adaptive collaborative search mechanism is used to collaboratively search the model parameter code after adaptive search to obtain the model parameter code after collaborative search, including: For any one dimension of the model parameter, the model parameters of all the model parameter codes after adaptive search are averaged, and the average values corresponding to all dimensions are constructed into a vector to obtain a center code; The model parameter codes after adaptive search are arranged in order of fitness from small to large to obtain the model parameter codes after arrangement; For the model parameter codes after arrangement, the first collaborative search factor and the second collaborative search factor are obtained as: wherein, represents the model parameter encoding corresponding to the first collaborative search factor after the i th permutation, represents the model parameter encoding corresponding to the second collaborative search factor after the i th permutation, sin represents a sine function, cos represents a cosine function, represents a collaborative search direction control parameter, and , represents a constant term, represents a third random number between (0, 1), represents a collaborative search range control parameter, and , represents a search range base influence parameter, represents a fourth random number between (0, 1); represents a search range base influence parameter, and According to the current training number, a position disturbance factor is obtained as: wherein, denotes a position perturbation factor, denotes the number pi, denotes a maximum value of the position perturbation factor, denotes a minimum value of the position perturbation factor, denotes a maximum number of training iterations; According to the center code, the position disturbance factor, the first collaborative search factor and the second collaborative search factor, the model parameter codes after arrangement are collaboratively searched to obtain the model parameter codes after collaborative search as: wherein, denotes the model parameter encoding after the i th permutation, i = 1, 2,..., K, K denotes the total number of model parameter encodings, denotes the model parameter encoding after the i + 1th permutation, and i = K, then is the model parameter encoding after a random permutation, denotes the model parameter encoding after the i th cooperative search, denotes the center encoding.

5. The artificial intelligence-based energy storage system energy management method of claim 4, wherein, An adaptive number of mutation fusion mechanisms is used to perform mutation fusion search on the model parameter codes after collaborative search to obtain the model parameter codes after mutation fusion search, including: According to the fitness of the model parameter codes after collaborative search, an adaptive mutation fusion number is obtained as: wherein, denotes the model parameter encoding corresponding to the adaptive mutation fusion number after the m th collaborative search, m = 1, 2,..., K, K denotes the total number of model parameter encodings, denotes the ceiling function, denotes the model parameter encoding corresponding to the fitness after the m th collaborative search, denotes the maximum fitness among all model parameter encodings, denotes the minimum fitness among all model parameter encodings; For the model parameter encoding after the first m cooperative search, randomly select one other cooperative search after the model parameter encoding, get one variation fusion encoding; According to the method The model parameter encoding after the mutation fusion search is obtained by performing a mutation fusion search on the model parameter encoding after the collaborative search. wherein, denotes the model parameter encoding after the m th cooperative search, denotes the model parameter encoding after the m th mutation fusion search, denotes the fifth random number between (0,1), denotes the model parameter encoding corresponding to the j th mutation fusion encoding, denotes the model parameter encoding corresponding to the maximum fitness.

6. The artificial intelligence-based energy storage system energy management method of claim 1, wherein, The real-time load influence sample of the energy supply area is collected, and the real-time load influence sample is identified by the electricity consumption prediction model to determine the predicted electricity consumption of the energy supply area, including: The real-time load influence sample of the energy supply area is collected; wherein the data structure of the real-time load influence sample is the same as that of the historical load influence sample; The real-time load influence sample is taken as the input of the electricity consumption prediction model to obtain the predicted electricity consumption of the energy supply area.

7. The artificial intelligence-based energy storage system energy management method of claim 6, wherein, The remaining electricity of the energy storage system is determined, and the corresponding charging power of the energy storage system is determined according to the remaining electricity and the predicted electricity consumption, including: The remaining electricity of the energy storage system is determined; The difference between the remaining electricity and the predicted electricity consumption is obtained as an electricity difference; In the case where the electricity difference is greater than or equal to zero, the corresponding charging power of the energy storage system is determined as a first preset charging power; In the case where the electricity difference is less than zero, a second preset charging power corresponding to the electricity difference is determined; wherein the second preset charging power is greater than the first preset charging power; different electricity differences have different second preset charging powers, and the electricity difference is proportional to the second preset charging power, and the first preset charging power and the second preset charging power are both less than or equal to the rated charging power designed for the energy storage system.

8. The artificial intelligence-based energy storage system energy management method of claim 7, wherein, Based on the corresponding charging power of the energy storage system, the energy management of the energy storage system is performed, including: Control the energy storage system to charge in a preset time period based on the charging power corresponding to the energy storage system, wherein the preset time period refers to a low electricity price period.

9. An artificial intelligence-based energy management apparatus for an energy storage system, the artificial intelligence-based energy management apparatus being capable of performing the artificial intelligence-based energy management method for an energy storage system according to any one of claims 1 to 8, characterized by Comprise: a data acquisition module, a data training module, an electricity consumption prediction module, a charging determination module, and an energy management module; The data acquisition module is configured to acquire historical electricity consumption correlation data corresponding to an energy supply area corresponding to an energy storage system, and to construct historical load influence samples and corresponding electricity consumption labels based on the historical electricity consumption correlation data; The data training module is configured to learn the data relationship between the historical load influence samples and the corresponding electricity consumption labels using an artificial intelligence model, and to obtain an electricity consumption prediction model; The electricity consumption prediction module is configured to acquire real-time load influence samples of the energy supply area, and to schedule the electricity consumption prediction model to identify the real-time load influence samples, and to determine the predicted electricity consumption of the energy supply area; The charging determination module is configured to determine the remaining electricity quantity of the energy storage system, and to determine the charging power corresponding to the energy storage system based on the remaining electricity quantity and the predicted electricity consumption; The energy management module is configured to perform energy management on the energy storage system based on the charging power corresponding to the energy storage system, thereby completing artificial intelligence-based energy management of the energy storage system.

Images