Power budget allocation method and device

Abstract

The present invention discloses a power budget allocation method and device, which belongs to the field of power supply technology. It is based on historical power consumption data, and after using a deep learning algorithm to learn the historical power consumption data, it can realize short-term power consumption prediction, and dynamically adjust the power allocation of each device in combination with the importance of the device, and reasonably allocate the power usage budget of different devices, so that the overall power usage is more efficient and power waste caused by excessive power consumption of individual devices is avoided.

Description

Technical Field

[0001] The present invention belongs to the technical field of power supply, and in particular relates to a power supply budget allocation method and device. Background Art

[0002] With the development of social economy, the problem of energy consumption is becoming increasingly serious. Power management systems are increasingly widely used in various types of equipment. How to reasonably allocate power budgets and improve power utilization efficiency has become an urgent problem to be solved. Power budgeting refers to the process of estimating and planning the power consumption of power systems or equipment under certain conditions. Most of the existing power budget allocation methods adopt average allocation or fixed allocation strategies, which cannot meet the differences in power requirements in different scenarios, resulting in power waste. Summary of the invention

[0003] The present invention provides a power budget allocation method and device to solve the problems existing in the prior art.

[0004] In one aspect, the present invention provides a power budget allocation method, comprising:

[0005] For each device connected to the power supply, historical power consumption data corresponding to the device is collected at a preset data sampling frequency, and sample data is constructed based on the historical power consumption data;

[0006] Using a neural network to build a short-term power consumption prediction model, and using the sample data to train the short-term power consumption prediction model to obtain a trained short-term power consumption prediction model;

[0007] For each device connected to the power supply, the current power consumption data of the device is collected and processed through the trained short-term power consumption prediction model to determine the predicted power consumption at a future time point.

[0008] Obtaining the sum of predicted power consumption corresponding to all devices, and determining the adequacy of power supply according to the sum of predicted power consumption; wherein the adequacy of supply includes sufficient supply or insufficient supply;

[0009] When the supply adequacy is sufficient, all devices are powered by predicted power consumption to achieve power budget allocation; when the supply adequacy is insufficient, some devices are powered by an importance round-robin algorithm to achieve power budget allocation.

[0010] Furthermore, for each device connected to the power supply, historical power consumption data corresponding to the device is collected at a preset data sampling frequency, and sample data is constructed with the historical power consumption data, including:

[0011] For each device connected to the power supply, the historical power consumption data corresponding to the device is collected at a preset data sampling frequency to obtain the actual power consumption of any device at multiple consecutive time points;

[0012] For the actual power consumption of any device at multiple consecutive time points, the actual power consumption at any N consecutive time points is used to construct a sample power consumption sequence, and the actual power consumption at the N+1th time point is used as the true value label corresponding to the sample power consumption sequence;

[0013] Sample data is obtained according to the sample power consumption sequence and the true value label corresponding to the sample power consumption sequence.

[0014] Further, the sample data is used to train the short-term power consumption prediction model to obtain the trained short-term power consumption prediction model, including:

[0015] Initializing the model parameters of the short-term power consumption prediction model, and determining a plurality of different model parameter vectors in the solution space; wherein the model parameter vector includes all parameters to be trained of the short-term power consumption prediction model;

[0016] Obtaining the fitness value corresponding to each model parameter vector, and determining the optimal parameter vector and the worst parameter vector according to the fitness value corresponding to each model parameter vector;

[0017] For each model parameter vector, the model parameter vector is fused with the optimal parameter vector to obtain the model parameter vector after information fusion;

[0018] For the model parameter vector after information fusion, the model parameter vector is subjected to double nonlinear information fusion with other model parameter vectors and the optimal parameter vector to obtain the model parameter vector after double nonlinear information fusion;

[0019] For the model parameter vector after the dual nonlinear information fusion, all the model parameter vectors are subjected to joint information fusion around the optimal parameter vector to obtain the model parameter vector after the joint information fusion;

[0020] For the model parameter vector after joint information fusion, mutually exclusive information fusion is performed according to the optimal parameter vector and the worst parameter vector to obtain the model parameter vector after mutually exclusive information fusion;

[0021] Determine whether the current number of training times has reached the maximum number of training times. If so, re-determine the optimal parameter vector based on the model parameter vector after mutually exclusive information fusion, and use the parameters contained in the optimal parameter vector as the final parameters of the short-term power consumption prediction model to obtain the short-term power consumption prediction model after training. Otherwise, return to the step of obtaining the fitness value.

[0022] Further, the fitness value corresponding to each model parameter vector is obtained, and the optimal parameter vector and the worst parameter vector are determined according to the fitness value corresponding to each model parameter vector, including:

[0023] Get the fitness value corresponding to each model parameter vector:

[0024]

[0025] Among them, f represents the fitness value, y n represents the actual output obtained by taking the nth sample power consumption sequence as input, Represents the true value label corresponding to the nth sample power consumption sequence;

[0026] The model parameter vector with the largest fitness value is determined as the optimal parameter vector, and the model parameter vector with the smallest fitness value is determined as the worst parameter vector.

[0027] Furthermore, for each model parameter vector, the model parameter vector is fused with the optimal parameter vector to obtain a model parameter vector after information fusion, including:

[0028] For each model parameter vector, according to the fitness value corresponding to the optimal parameter vector, the information interaction coefficient corresponding to the model parameter vector is determined as:

[0029] ω i =0.5+exp(-f i / f best ) t

[0030] Among them, ω i represents the information interaction coefficient corresponding to the i-th model parameter vector, f i represents the fitness corresponding to the i-th model parameter vector, f best represents the fitness value corresponding to the optimal parameter vector, and t represents the current number of training times;

[0031] According to the information interaction coefficient corresponding to the model parameter vector, the model parameter vector and the optimal parameter vector are information-fused as follows:

[0032]

[0033] in, represents the optimal parameter vector, represents the i-th model parameter vector in the t-th training process, Represents the model parameter vector after information fusion I represents the total number of model parameter vectors, α represents the update control factor, and r1 represents a random number between (0,1), T represents the maximum number of training times, and c represents a coefficient between (0,2).

[0034] Furthermore, for the model parameter vector after information fusion, the model parameter vector is subjected to double nonlinear information fusion with other model parameter vectors and the optimal parameter vector to obtain the model parameter vector after double nonlinear information fusion, including:

[0035] For the model parameter vector after information fusion, re-obtain the fitness value of each model parameter vector, and arrange them in descending order of fitness value to obtain the model parameter vector after the first sorting;

[0036] For the model parameter vector after the first sorting, the model parameter vector is subjected to double nonlinear information fusion with other model parameter vectors and the optimal parameter vector as follows:

[0037]

[0038] in, represents the model parameter vector after the jth first sort in the tth training process, ω j Represents the model parameter vector The corresponding information interaction coefficient is, represents the optimal parameter vector, Represents the model parameter vector after dual nonlinear information fusion represents the model parameter vector after the j-1th first sort in the tth training process, e represents a natural constant, r2 represents a random number between (0,1), r3 represents a random number between (-1,1), and π represents the ratio of pi.

[0039] Furthermore, for the model parameter vector after the dual nonlinear information fusion, all model parameter vectors are subjected to joint information fusion around the optimal parameter vector to obtain the model parameter vector after the joint information fusion, including:

[0040] For the model parameter vector after the dual nonlinear information fusion, re-obtain the fitness value of each model parameter vector, and arrange them in descending order according to the fitness value to obtain the model parameter vector after the second sorting;

[0041] For the model parameter vector after the second sorting, all model parameter vectors are subjected to joint information fusion around the optimal parameter vector as follows:

[0042]

[0043] in, represents the model parameter vector after the mth second sort in the tth training process, Represents the model parameter vector after joint information fusion represents the model parameter vector after the m-1th second sorting in the tth training process, β represents the adaptive learning coefficient, r4 represents a random number between (0,1), r5 represents a random number between (0,1), and T represents the maximum number of training times.

[0044] Furthermore, for the model parameter vector after joint information fusion, mutually exclusive information fusion is performed according to the optimal parameter vector and the worst parameter vector, and the model parameter vector after mutually exclusive information fusion is obtained as follows:

[0045]

[0046] in, represents the model parameter vector after the kth joint information fusion in the tth training process, Represents the model parameter vector after mutually exclusive information fusion r6 represents a random number between (0,1), represents the worst parameter vector.

[0047] Furthermore, when the supply adequacy is insufficient, an importance round-robin algorithm is used to supply power to some devices, including:

[0048] When the supply adequacy is insufficient, a preset importance corresponding to each device is rotated;

[0049] According to the preset importance of the equipment, determine whether the sum of the predicted power consumption corresponding to the first M devices is less than the output power of the power supply. If so, power is supplied to the first M devices, and other devices are powered one by one according to the preset minimum working power until the power supply can no longer provide power to more devices. Otherwise, starting from the Mth device, devices are eliminated one by one until the power supply is normally supplied.

[0050] On the other hand, the present invention provides a power budget allocation device, including: a data acquisition module, a model building module, a prediction module, a judgment module and a power supply budget allocation module;

[0051] The data acquisition module is used to collect historical power consumption data corresponding to each device connected to the power supply according to a preset data sampling frequency, and construct sample data with the historical power consumption data;

[0052] The model building module is used to use a neural network to build a short-term power consumption prediction model, and use the sample data to train the short-term power consumption prediction model to obtain a trained short-term power consumption prediction model;

[0053] The prediction module is used to collect current power consumption data corresponding to each device connected to the power supply, and process the current power consumption data through the trained short-term power consumption prediction model to determine the predicted power consumption at a future time point;

[0054] The judgment module is used to obtain the sum of the predicted power consumption corresponding to all devices, and determine the adequacy of power supply according to the sum of the predicted power consumption; wherein the adequacy of supply includes sufficient supply or insufficient supply;

[0055] The power supply budget allocation module is used to supply power to all devices based on predicted power consumption to achieve power budget allocation when the supply adequacy is sufficient; when the supply adequacy is insufficient, an importance round-robin algorithm is used to supply power to some devices to achieve power budget allocation.

[0056] The present invention provides a power budget allocation method and device, which first uses historical power consumption data as a basis, and adopts a deep learning algorithm to learn the historical power consumption data, and then can achieve short-term power consumption prediction, and dynamically adjust the power allocation of each device in combination with the importance of the device, and reasonably allocate the power usage budget of different devices, so that the overall power usage is more efficient and avoids power waste caused by excessive power consumption of individual devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] 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.

[0058] Figure 1 A flow chart of a power budget allocation method provided by an embodiment of the present invention.

[0059] Figure 2 A schematic diagram of the structure of a power budget allocation device provided in an embodiment of the present invention.

[0060] 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

[0061] 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.

[0062] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0063] like Figure 1 As shown, an embodiment of the present invention provides a power budget allocation method, including:

[0064] S1. For each device connected to a power source, collect historical power consumption data corresponding to the device according to a preset data sampling frequency, and construct sample data based on the historical power consumption data;

[0065] For a certain type of automatically running equipment, the power consumption during use has certain regularities. Therefore, historical power consumption data can be used and studied to explore the operating regularities of the equipment, thereby realizing power budget allocation.

[0066] It is worth noting that for manually operated devices, the operation of such devices is limitedly guaranteed, and the power budget proposed in the embodiment of the present invention is only for automatically operated devices.

[0067] S2. Using a neural network to construct a short-term power consumption prediction model, and using the sample data to train the short-term power consumption prediction model to obtain a trained short-term power consumption prediction model;

[0068] For example, a BP (Back Propagatio) neural network or an LSTM (Long Short-Term Memory) neural network can be used to build a short-term power consumption prediction model. After the short-term power consumption prediction model is built, an optimization algorithm can be used to optimize its hyperparameters to obtain a short-term power consumption prediction model with short-term power consumption prediction capability.

[0069] It is worth noting that in order to ensure the accuracy of short-term power consumption prediction, other data can also be collected to construct sample data.

[0070] S3. For each device connected to the power supply, collect the current power consumption data corresponding to the device, and process the current power consumption data through the trained short-term power consumption prediction model to determine the predicted power consumption at a future time point;

[0071] The current power consumption data refers to the power consumption collected at N time points forward starting from the current time, and the power consumption at these N time points is used as input data to construct a power consumption short-term prediction model, so as to determine the predicted power consumption at the next time point.

[0072] S4. Obtain the sum of the predicted power consumption corresponding to all devices, and determine the adequacy of power supply according to the sum of the predicted power consumption; wherein the adequacy of supply includes sufficient supply or insufficient supply;

[0073] When the sum of the predicted power consumption is less than the rated output power consumption of the power supply, it can be considered that the power supply is sufficient, so the power budget can be allocated to each device according to the predicted power consumption.

[0074] When the sum of the predicted power consumption is greater than the rated output power consumption of the power supply, it can be considered that the power supply is insufficient, so it is necessary to allocate the power budget based on the importance of the equipment.

[0075] S5. When the supply adequacy is sufficient, all devices are powered according to the predicted power consumption to achieve power budget allocation; when the supply adequacy is insufficient, some devices are powered by an importance round-robin algorithm to achieve power budget allocation.

[0076] The present invention provides a power budget allocation method, which is based on historical power consumption data. After using a deep learning algorithm to learn the historical power consumption data, it can achieve short-term power consumption prediction, and dynamically adjust the power allocation of each device in combination with the importance of the device, and reasonably allocate the power usage budget of different devices, so that the overall power usage is more efficient and power waste caused by excessive power consumption of individual devices is avoided.

[0077] In an embodiment of the present invention, for each device connected to a power source, historical power consumption data corresponding to the device is collected at a preset data sampling frequency, and sample data is constructed using the historical power consumption data, including:

[0078] For each device connected to the power supply, the historical power consumption data corresponding to the device is collected at a preset data sampling frequency to obtain the actual power consumption of any device at multiple consecutive time points;

[0079] For the actual power consumption of any device at multiple consecutive time points, the actual power consumption at any N consecutive time points is used to construct a sample power consumption sequence, and the actual power consumption at the N+1th time point is used as the true value label corresponding to the sample power consumption sequence;

[0080] Sample data is obtained according to the sample power consumption sequence and the true value label corresponding to the sample power consumption sequence.

[0081] In an embodiment of the present invention, the sample data is used to train the short-term power consumption prediction model to obtain the trained short-term power consumption prediction model, including:

[0082] Initializing the model parameters of the short-term power consumption prediction model, and determining a plurality of different model parameter vectors in the solution space; wherein the model parameter vector includes all parameters to be trained of the short-term power consumption prediction model;

[0083] For example, when using BP neural network to build a short-term power consumption prediction model, its hyperparameters are mainly weight parameters, and each weight parameter has corresponding upper and lower limits. The training process of BP neural network is essentially to find the best combination of these weight parameters. Therefore, the weight parameters can be randomly initialized between the upper and lower limits, and the weight parameters after initialization constitute the model parameter vector.

[0084] Obtaining the fitness value corresponding to each model parameter vector, and determining the optimal parameter vector and the worst parameter vector according to the fitness value corresponding to each model parameter vector;

[0085] For each model parameter vector, the model parameter vector is fused with the optimal parameter vector to obtain the model parameter vector after information fusion;

[0086] For the model parameter vector after information fusion, the model parameter vector is subjected to double nonlinear information fusion with other model parameter vectors and the optimal parameter vector to obtain the model parameter vector after double nonlinear information fusion;

[0087] For the model parameter vector after the dual nonlinear information fusion, all the model parameter vectors are subjected to joint information fusion around the optimal parameter vector to obtain the model parameter vector after the joint information fusion;

[0088] For the model parameter vector after joint information fusion, mutually exclusive information fusion is performed according to the optimal parameter vector and the worst parameter vector to obtain the model parameter vector after mutually exclusive information fusion;

[0089] Determine whether the current number of training times has reached the maximum number of training times. If so, re-determine the optimal parameter vector based on the model parameter vector after mutually exclusive information fusion, and use the parameters contained in the optimal parameter vector as the final parameters of the short-term power consumption prediction model to obtain the short-term power consumption prediction model after training. Otherwise, return to the step of obtaining the fitness value.

[0090] In the prior art, although genetic algorithms have strong randomness in the process of training neural networks, they have poor convergence and slow search speed, and it is difficult to find the global optimal solution. Therefore, the embodiment of the present invention provides an optimization algorithm to solve the technical problems existing in the prior art.

[0091] In an embodiment of the present invention, obtaining the fitness value corresponding to each model parameter vector, and determining the optimal parameter vector and the worst parameter vector according to the fitness value corresponding to each model parameter vector, includes:

[0092] Get the fitness value corresponding to each model parameter vector:

[0093]

[0094] Among them, f represents the fitness value, y n represents the actual output obtained by taking the nth sample power consumption sequence as input, Represents the true value label corresponding to the nth sample power consumption sequence;

[0095] The model parameter vector with the largest fitness value is determined as the optimal parameter vector, and the model parameter vector with the smallest fitness value is determined as the worst parameter vector.

[0096] It is worth noting that the denominator of the fitness value can be added to a very small constant (such as 0.001) to avoid the denominator being zero.

[0097] In the embodiment of the present invention, for each model parameter vector, information fusion is performed on the model parameter vector and the optimal parameter vector to obtain the model parameter vector after information fusion, including:

[0098] For each model parameter vector, according to the fitness value corresponding to the optimal parameter vector, the information interaction coefficient corresponding to the model parameter vector is determined as:

[0099] ω i =0.5+exp(-f i / f best ) t

[0100] Among them, ω i represents the information interaction coefficient corresponding to the i-th model parameter vector, f i represents the fitness corresponding to the i-th model parameter vector, f best represents the fitness value corresponding to the optimal parameter vector, and t represents the current number of training times;

[0101] According to the information interaction coefficient corresponding to the model parameter vector, the model parameter vector and the optimal parameter vector are information-fused as follows:

[0102]

[0103] in, represents the optimal parameter vector, represents the i-th model parameter vector in the t-th training process, Represents the model parameter vector after information fusion I represents the total number of model parameter vectors, α represents the update control factor, and r1 represents a random number between (0,1), T represents the maximum number of training times, and c represents a coefficient between (0,2).

[0104] The information fusion method proposed in the embodiment of the present invention can fuse the model parameter vector with the optimal parameter vector, so as to explore the area between the optimal position and advance with an adaptive step size, thereby ensuring both the search speed in the early stage of the algorithm and the search accuracy in the later stage of the algorithm.

[0105] In an embodiment of the present invention, for a model parameter vector after information fusion, the model parameter vector is subjected to double nonlinear information fusion with other model parameter vectors and an optimal parameter vector to obtain a model parameter vector after double nonlinear information fusion, including:

[0106] For the model parameter vector after information fusion, re-obtain the fitness value of each model parameter vector, and arrange them in descending order of fitness value to obtain the model parameter vector after the first sorting;

[0107] For the model parameter vector after the first sorting, the model parameter vector is subjected to double nonlinear information fusion with other model parameter vectors and the optimal parameter vector as follows:

[0108]

[0109] in, represents the model parameter vector after the jth first sort in the tth training process, ω j Represents the model parameter vector The corresponding information interaction coefficient is, represents the optimal parameter vector, Represents the model parameter vector after dual nonlinear information fusion represents the model parameter vector after the j-1th first sort in the tth training process, e represents a natural constant, r2 represents a random number between (0,1), r3 represents a random number between (-1,1), and π represents the ratio of pi.

[0110] The dual nonlinear information fusion provided by the embodiment of the present invention can enable the model parameter vector to learn information of a better position and information of the optimal position, thereby effectively improving the search speed of the algorithm.

[0111] In the embodiment of the present invention, for the model parameter vector after the dual nonlinear information fusion, all the model parameter vectors are subjected to joint information fusion around the optimal parameter vector to obtain the model parameter vector after the joint information fusion, including:

[0112] For the model parameter vector after the dual nonlinear information fusion, re-obtain the fitness value of each model parameter vector, and arrange them in descending order according to the fitness value to obtain the model parameter vector after the second sorting;

[0113] For the model parameter vector after the second sorting, all model parameter vectors are subjected to joint information fusion around the optimal parameter vector as follows:

[0114]

[0115] in, represents the model parameter vector after the mth second sort in the tth training process, Represents the model parameter vector after joint information fusion represents the model parameter vector after the m-1th second sorting in the tth training process, β represents the adaptive learning coefficient, r4 represents a random number between (0,1), r5 represents a random number between (0,1), and T represents the maximum number of training times.

[0116] The joint information fusion provided by the embodiment of the present invention enables all model parameter vectors to perform collaborative search around the optimal position, provides a stronger search capability for the optimal area, and can effectively find the global optimal value.

[0117] In the embodiment of the present invention, for the model parameter vector after the joint information fusion, mutually exclusive information fusion is performed according to the optimal parameter vector and the worst parameter vector, and the model parameter vector after the mutually exclusive information fusion is obtained as follows:

[0118]

[0119] in, represents the model parameter vector after the kth joint information fusion in the tth training process, Represents the model parameter vector after mutually exclusive information fusion r6 represents a random number between (0,1), represents the worst parameter vector.

[0120] The mutually exclusive information fusion provided by the embodiment of the present invention can effectively enhance the global search capability of the algorithm, improve the diversity of the algorithm, and effectively prevent the algorithm from falling into a local optimum.

[0121] Optionally, the greedy principle can be used to control the mutually exclusive information fusion to ensure the search speed of the algorithm. It is also possible to perform out-of-bounds processing on the model parameter vector after it changes.

[0122] In the embodiment of the present invention, when the supply adequacy is insufficient supply, an importance round-robin algorithm is used to supply power to some devices, including:

[0123] When the supply adequacy is insufficient, a preset importance corresponding to each device is rotated;

[0124] According to the preset importance of the equipment, determine whether the sum of the predicted power consumption corresponding to the first M devices is less than the output power of the power supply. If so, power is supplied to the first M devices, and other devices are powered one by one according to the preset minimum working power until the power supply can no longer provide power to more devices (that is, adding power to one more device will cause the power supply to be overloaded). Otherwise, starting from the Mth device, devices are eliminated one by one until the power supply is normally supplied.

[0125] like Figure 2 As shown, an embodiment of the present invention provides a power budget allocation device, including: a data acquisition module 1, a model building module 2, a prediction module 3, a judgment module 4 and a power supply budget allocation module 5;

[0126] The data acquisition module 1 is used to collect historical power consumption data corresponding to each device connected to the power supply according to a preset data sampling frequency, and construct sample data based on the historical power consumption data;

[0127] The model building module 2 is used to build a short-term power consumption prediction model using a neural network, and train the short-term power consumption prediction model using the sample data to obtain a trained short-term power consumption prediction model;

[0128] The prediction module 3 is used to collect the current power consumption data corresponding to each device connected to the power supply, and process the current power consumption data through the trained short-term power consumption prediction model to determine the predicted power consumption at a future time point;

[0129] The judgment module 4 is used to obtain the sum of the predicted power consumption corresponding to all devices, and determine the adequacy of power supply according to the sum of the predicted power consumption; wherein the adequacy of supply includes sufficient supply or insufficient supply;

[0130] The power supply budget allocation module 5 is used to supply power to all devices based on predicted power consumption to achieve power budget allocation when the supply adequacy is sufficient; when the supply adequacy is insufficient, an importance round-robin algorithm is used to supply power to some devices to achieve power budget allocation.

[0131] A power budget allocation device provided in an embodiment of the present invention can execute the above method and technical solution, and its principles and beneficial effects are similar, which will not be described in detail here.

[0132] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0133] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0134] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0135] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0136] A person of ordinary skill in the art can understand that all or part of the steps in realizing the above-mentioned facts and methods can be completed by instructing the relevant hardware through a program, and the program involved or the program described can be stored in a computer-readable storage medium. When the program is executed, it includes the following steps: At this time, the corresponding method steps are derived, and the storage medium can be ROM / RAM, a disk, an optical disk, etc.

[0137] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A power budget allocation method, characterized in that: include: For each device connected to the power supply, historical power consumption data corresponding to the device is collected at a preset data sampling frequency, and sample data is constructed based on the historical power consumption data; Using a neural network to build a short-term power consumption prediction model, and using the sample data to train the short-term power consumption prediction model to obtain a trained short-term power consumption prediction model; For each device connected to the power supply, the current power consumption data of the device is collected and processed through the trained short-term power consumption prediction model to determine the predicted power consumption at a future time point. Obtaining the sum of predicted power consumption corresponding to all devices, and determining the adequacy of power supply according to the sum of predicted power consumption; wherein the adequacy of supply includes sufficient supply or insufficient supply; When the supply adequacy is sufficient, power is supplied to all devices according to the predicted power consumption to realize power budget allocation; when the supply adequacy is insufficient, power is supplied to some devices using the importance round-robin algorithm to realize power budget allocation; For each device connected to the power supply, historical power consumption data corresponding to the device is collected at a preset data sampling frequency, and sample data is constructed based on the historical power consumption data, including: For each device connected to the power supply, the historical power consumption data corresponding to the device is collected at a preset data sampling frequency to obtain the actual power consumption of any device at multiple consecutive time points; For the actual power consumption of any device at multiple consecutive time points, the actual power consumption at any N consecutive time points is used to construct a sample power consumption sequence, and the actual power consumption at the N+1th time point is used as the true value label corresponding to the sample power consumption sequence; Obtain sample data according to the sample power consumption sequence and the true value label corresponding to the sample power consumption sequence; The sample data is used to train the short-term power consumption prediction model to obtain the trained short-term power consumption prediction model, including: Initializing the model parameters of the short-term power consumption prediction model, and determining a plurality of different model parameter vectors in the solution space; wherein the model parameter vector includes all parameters to be trained of the short-term power consumption prediction model; Obtaining the fitness value corresponding to each model parameter vector, and determining the optimal parameter vector and the worst parameter vector according to the fitness value corresponding to each model parameter vector; For each model parameter vector, the model parameter vector is fused with the optimal parameter vector to obtain the model parameter vector after information fusion; For the model parameter vector after information fusion, the model parameter vector is subjected to double nonlinear information fusion with other model parameter vectors and the optimal parameter vector to obtain the model parameter vector after double nonlinear information fusion; For the model parameter vector after the dual nonlinear information fusion, all the model parameter vectors are subjected to joint information fusion around the optimal parameter vector to obtain the model parameter vector after the joint information fusion; For the model parameter vector after joint information fusion, mutually exclusive information fusion is performed according to the optimal parameter vector and the worst parameter vector to obtain the model parameter vector after mutually exclusive information fusion; Determine whether the current number of training times has reached the maximum number of training times. If so, re-determine the optimal parameter vector based on the model parameter vector after mutually exclusive information fusion, and use the parameters contained in the optimal parameter vector as the final parameters of the short-term power consumption prediction model to obtain the short-term power consumption prediction model after training. Otherwise, return to the step of obtaining the fitness value.

2. The power budget allocation method according to claim 1, characterized in that: Obtain the fitness value corresponding to each model parameter vector, and determine the optimal parameter vector and the worst parameter vector according to the fitness value corresponding to each model parameter vector, including: Get the fitness value corresponding to each model parameter vector: in, f represents the fitness value, represents the actual output obtained by taking the nth sample power consumption sequence as input, Represents the true value label corresponding to the nth sample power consumption sequence; The model parameter vector with the largest fitness value is determined as the optimal parameter vector, and the model parameter vector with the smallest fitness value is determined as the worst parameter vector.

3. The power budget allocation method according to claim 2, characterized in that: For each model parameter vector, the model parameter vector is fused with the optimal parameter vector to obtain the model parameter vector after information fusion, including: For each model parameter vector, according to the fitness value corresponding to the optimal parameter vector, the information interaction coefficient corresponding to the model parameter vector is determined as: in, Indicates i The information interaction coefficient corresponding to the model parameter vector is Indicates i The fitness corresponding to the model parameter vector is represents the fitness value corresponding to the optimal parameter vector, and t represents the current number of training times; According to the information interaction coefficient corresponding to the model parameter vector, the model parameter vector and the optimal parameter vector are information-fused as follows: in, represents the optimal parameter vector, Indicates the tth training process i A model parameter vector, Represents the model parameter vector after information fusion , i =1,2,…,I, where I represents the total number of model parameter vectors, represents the update control factor, and , represents a random number between (0,1), T represents the maximum number of training times, c Represents a coefficient between (0,2).

4. The power budget allocation method according to claim 3, characterized in that: For the model parameter vector after information fusion, the model parameter vector is subjected to double nonlinear information fusion with other model parameter vectors and the optimal parameter vector to obtain the model parameter vector after double nonlinear information fusion, including: For the model parameter vector after information fusion, re-obtain the fitness value of each model parameter vector, and arrange them in descending order of fitness value to obtain the model parameter vector after the first sorting; For the model parameter vector after the first sorting, the model parameter vector is subjected to double nonlinear information fusion with other model parameter vectors and the optimal parameter vector as follows: in, Indicates the tth training process j The model parameter vector after the first sorting, Represents the model parameter vector The corresponding information interaction coefficient is, represents the optimal parameter vector, represents the model parameter vector after dual nonlinear information fusion, Indicates the tth training process j -1 model parameter vector after the first sort, e represents a natural constant, Represents a random number between (0,1), Represents a random number between (-1,1), Represents pi.

5. The power budget allocation method according to claim 4, characterized in that: For the model parameter vector after the dual nonlinear information fusion, all model parameter vectors are subjected to joint information fusion around the optimal parameter vector to obtain the model parameter vector after the joint information fusion, including: For the model parameter vector after the dual nonlinear information fusion, re-obtain the fitness value of each model parameter vector, and arrange them in descending order according to the fitness value to obtain the model parameter vector after the second sorting; For the model parameter vector after the second sorting, all model parameter vectors are subjected to joint information fusion around the optimal parameter vector as follows: in, Indicates t During the training process m The model parameter vector after the second sort, represents the model parameter vector after joint information fusion, Indicates t During the training process m -1 model parameter vector after the second sort, represents the adaptive learning coefficient, Represents a random number between (0,1), represents a random number between (0,1), and T represents the maximum number of training times.

6. The power budget allocation method according to claim 5, characterized in that: For the model parameter vector after joint information fusion, mutually exclusive information fusion is performed according to the optimal parameter vector and the worst parameter vector, and the model parameter vector after mutually exclusive information fusion is obtained as follows: in, Indicates the tth training process k The model parameter vector after joint information fusion, represents the model parameter vector after mutually exclusive information fusion, Represents a random number between (0,1), represents the worst parameter vector.

7. The power budget allocation method according to claim 1, characterized in that: When the supply adequacy is insufficient, an importance round-robin algorithm is used to supply power to some devices, including: When the supply adequacy is insufficient, a preset importance corresponding to each device is rotated; According to the preset importance of the equipment, determine whether the sum of the predicted power consumption corresponding to the first M devices is less than the output power of the power supply. If so, power is supplied to the first M devices, and other devices are powered one by one according to the preset minimum working power until the power supply can no longer provide power to more devices. Otherwise, starting from the Mth device, devices are eliminated one by one until the power supply is normally supplied.

8. A power budget allocation device, the power budget allocation device being used to execute the power budget allocation method according to any one of claims 1 to 7, characterized in that: include: Data acquisition module, model building module, prediction module, judgment module and power supply budget allocation module; The data acquisition module is used to collect historical power consumption data corresponding to each device connected to the power supply according to a preset data sampling frequency, and construct sample data with the historical power consumption data; The model building module is used to use a neural network to build a short-term power consumption prediction model, and use the sample data to train the short-term power consumption prediction model to obtain a trained short-term power consumption prediction model; The prediction module is used to collect current power consumption data corresponding to each device connected to the power supply, and process the current power consumption data through the trained short-term power consumption prediction model to determine the predicted power consumption at a future time point; The judgment module is used to obtain the sum of the predicted power consumption corresponding to all devices, and determine the adequacy of power supply according to the sum of the predicted power consumption; wherein the adequacy of supply includes sufficient supply or insufficient supply; The power supply budget allocation module is used to supply power to all devices based on predicted power consumption to achieve power budget allocation when the supply adequacy is sufficient; when the supply adequacy is insufficient, an importance round-robin algorithm is used to supply power to some devices to achieve power budget allocation.

Images