Intelligent management system of optical storage equipment

Abstract

The invention relates to the field of optical storage equipment management, in particular to an intelligent management system of optical storage equipment, which comprises photovoltaic equipment, a photovoltaic inverter, an energy storage system, a load, a power distribution network, a transformer and an optical storage management platform, and is characterized in that the optical storage management platform comprises a data acquisition module, a performance prediction module, an optimization scheduling module and a fault diagnosis module. According to the method, a fusion model based on multiple data is designed, local information features are extracted from multiple dimensions through multiple deep learning algorithms, then the three features are effectively fused for deep learning and prediction, and high-precision prediction of the performance of the optical storage system is achieved; according to the invention, the improved particle swarm optimization algorithm is used to carry out optimal scheduling on the charging and discharging of the photovoltaic power generation and energy storage system, a minimum economic cost objective function and multiple constraint conditions are designed, and the kNN algorithm is combined to improve the speed updating method of the particle swarm. And the economic cost of operation is obviously reduced while the supply and demand balance of the optical storage system is ensured.

Description

Technical Field

[0001] The present invention relates to the field of optical storage equipment management, and in particular to an intelligent management system for optical storage equipment. Background Art

[0002] With the rapid development of the economy, energy demand continues to grow. As a clean and renewable energy source, solar energy is gaining more and more attention. As an important part of the solar power generation system, the operating efficiency and management level of photovoltaic storage equipment directly affect the performance of the entire power grid. However, the existing intelligent management system of photovoltaic storage equipment has the problem that the prediction model only focuses on the dependency of time series, and it is difficult to fully extract the characteristic information of the data, resulting in inaccurate prediction results; when using the optimization algorithm to schedule the energy of the nodes of the photovoltaic storage system, it fails to effectively combine the prediction results of photovoltaic power generation and load demand, resulting in the scheduling plan being unable to adapt to the actual photovoltaic power generation fluctuations and load changes. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an intelligent management system for photovoltaic storage equipment. In view of the problem that the existing prediction model only focuses on the dependency of time series and it is difficult to fully extract the characteristic information of the data, resulting in inaccurate prediction results, the present invention designs a fusion model based on multiple data, combines weather characteristics and periodic characteristics, and predicts photovoltaic power generation, load power consumption and load demand through a variety of deep learning algorithms. Specifically, the multi-data fusion model deeply mines the long-term dependencies in the time series data through the LSTM model, captures the bidirectional context information features through the Bi-LSTM model, and finely extracts the local information features through the convolution module. By effectively fusing these three features and further performing deep learning and prediction, a high-precision prediction of the performance of the photovoltaic storage system is achieved; in view of the problem that when using the optimization algorithm to schedule the energy of the nodes of the photovoltaic storage system, the prediction results of photovoltaic power generation and load demand are not effectively combined, resulting in the scheduling scheme being unable to adapt to the actual photovoltaic power generation fluctuations and load changes, the present invention optimizes the charging and discharging of photovoltaic power generation and energy storage systems through an improved particle swarm optimization algorithm, combines the prediction results of photovoltaic power generation and load demand while fully considering the possible errors in the prediction results, designs the minimization of economic cost objective function and multiple constraints, and improves the speed update formula of the particle swarm algorithm in combination with the kNN algorithm, which significantly reduces the economic cost of operation while ensuring the supply and demand balance of the photovoltaic storage system.

[0004] The technical solution adopted by the present invention is as follows: an intelligent management system for photovoltaic storage equipment, specifically including photovoltaic equipment, photovoltaic inverters, energy storage systems, loads, distribution networks, transformers and a photovoltaic storage management platform, the photovoltaic storage management platform specifically including a data acquisition module, a performance prediction module, an optimization scheduling module and a fault diagnosis module;

[0005] The data acquisition module obtains the output power time series data of each photovoltaic device, the input power time series data of the load and the local weather time series data;

[0006] The performance prediction module constructs two fusion models based on multiple data with consistent structures, respectively performs photovoltaic power generation prediction and load power consumption prediction, and then calculates the load demand prediction;

[0007] The optimization scheduling module uses the power generation output power of the photovoltaic equipment, the discharge output power of the energy storage system and the charging output power of the energy storage system as decision variables, combines the results of photovoltaic power generation prediction and load demand prediction, defines the objective function and constraint conditions, and uses the improved particle swarm algorithm to optimize the power generation of each photovoltaic equipment and the charging and discharging of the energy storage system.

[0008] The fault diagnosis module monitors various system parameters in real time, sets a normal working range, and automatically issues an alarm when a parameter exceeds the normal working range;

[0009] The performance prediction module constructs two fusion models based on multiple data with the same structure, respectively performs photovoltaic power generation prediction and load power consumption prediction, and then calculates the load demand prediction process, which includes the following steps:

[0010] Step S1: convert the output power time series data of each photovoltaic device into an embedding vector to obtain photovoltaic power generation embedding, and convert the input power time series data of the load into an embedding vector to obtain load power embedding;

[0011] Step S2: convert each type of data in the weather time series data into an embedded vector in turn and concatenate them, and perform linear transformation on the concatenated results to obtain weather features;

[0012] Step S3: extract the date part of the output power time series data of each photovoltaic device, retain the month and day of the date, convert the date into an embedded vector, and obtain the periodic feature;

[0013] Step S4: construct a first fusion model based on multiple data and a second fusion model based on multiple data, wherein the structures of the first fusion model based on multiple data and the second fusion model based on multiple data are consistent, and the first fusion model based on multiple data specifically includes an input layer, a first LSTM layer, a second LSTM layer, a Bi-LSTM layer, a first Conv1d layer, a second Conv1d layer, a maximum pooling layer, a first dense layer, and a second dense layer;

[0014] Step S5: embedding photovoltaic power generation, weather characteristics and periodic characteristics are spliced ​​and put into the first multi-data based fusion model to obtain the photovoltaic power generation prediction result of each photovoltaic device;

[0015] Step S6: embedding the load power consumption, weather characteristics and period characteristics, and splicing them into the second multi-data based fusion model to obtain the load power consumption prediction result;

[0016] Step S7: Add the photovoltaic power generation prediction results of each photovoltaic device and subtract the load power consumption prediction result to obtain the load demand prediction result.

[0017] The specific steps of constructing the first multi-data based fusion model in step S4 are as follows:

[0018] Step S41: taking the data of the input layer as an input sequence;

[0019] Step S42: inputting the input sequence into the first LSTM layer, and inputting the output result of the first LSTM layer into the second LSTM layer to obtain LSTM features; the first LSTM layer and the second LSTM layer each include an LSTM model and a Dropout layer connected in sequence in a unidirectional manner;

[0020] Step S43: Input the input sequence to the Bi-LSTM layer to obtain Bi-LSTM features. The Bi-LSTM layer includes a Bi-LSTM model and a Dropout layer that are unidirectionally connected in sequence.

[0021] Step S44: Input the input sequence to the first Convld layer, input the output result of the first Convld layer to the second Convld layer, and input the output result of the second Convld layer to the maximum pooling layer to obtain convolution features. The specific formula is as follows:

[0022] The specific formula of the convolution operation of the Conv1d layer is as follows:

[0023]

[0024] Among them, yq,w represents the output of the convolution operation of the (q, w) coordinates, M and K represent the maximum values ​​of the horizontal and vertical coordinates respectively, and X q+m，w+n Data representing the (q+m, w+n) coordinates of the input data, F m，n Represents the value of the filter at the (m, n) coordinate.

[0025] Step S45: concatenate the LSTM features, the Bi-LSTM features and the convolution features to obtain fusion features;

[0026] Step S46: Input the fused features into the first dense layer, and input the output result of the first dense layer into the second dense layer to obtain the output result of the first multi-data based fusion model.

[0027] The optimization scheduling module takes the power generation output power of the photovoltaic equipment, the discharge output power of the energy storage system and the charging output power of the energy storage system as decision variables, combines the results of photovoltaic power generation prediction and load demand prediction, defines the objective function and constraint conditions, and uses the improved particle swarm algorithm to optimize the power generation of each photovoltaic equipment and the charging and discharging of the energy storage system. The specific steps of the scheduling process are as follows:

[0028] Step D1: Define the objective function of minimizing economic cost. The specific formula is as follows:

[0029]

[0030] Where min C represents the minimum economic cost, T represents the maximum time period, N represents the number of photovoltaic devices, and P i，t represents the power output of the ith photovoltaic device at time t, Indicates P i，t The square of a i 、b i 、c i are coefficients of variation, u i，t is the economic cost coefficient.

[0031] Step D2: Define constraints. Constraints include supply and demand balance constraints, energy storage system charging and discharging constraints, and photovoltaic equipment power generation constraints. The specific formula is as follows:

[0032] The specific formula of the supply and demand balance constraint is as follows:

[0033]

[0034] in, represents the discharge output power of the energy storage system at time t, represents the charging output power of the energy storage system at time t, It represents the load demand forecast result at the tth moment.

[0035] The specific formula for the energy storage system charging and discharging constraints is as follows:

[0036]

[0037]

[0038] in, Indicates the maximum limit of the discharge output power of the energy storage system, Indicates the maximum limit of the charging output power of the energy storage system.

[0039] The specific formula of the photovoltaic equipment power generation constraint is as follows:

[0040]

[0041] 0≤P i，t ≤P max ;

[0042] in, represents the photovoltaic power generation prediction result of the i-th photovoltaic device at time t, δ is the deviation coefficient, P max is the maximum output power of the photovoltaic device.

[0043] Step D3: Use the improved particle swarm algorithm to optimize the scheduling of power generation of each photovoltaic device and charging and discharging of the energy storage system. The specific steps are as follows:

[0044] Step D31: define the maximum number of iterations and initialize the particle swarm;

[0045] Step D32: taking the minimized economic cost function as the fitness function and calculating the fitness of each particle;

[0046] Step D33: Update the optimal solution of the particle swarm and each particle according to the fitness. The smaller the fitness, the better the particle.

[0047] Step D34: Use the kNN algorithm to find the neighboring solution set of each particle and update the velocity and position of the particle. The specific formula is as follows:

[0048] The particle velocity update formula is as follows:

[0049]

[0050] in, and denote the speed of the p-th particle at the o+1th and oth iterations, ω o represents the inertia weight of the oth iteration, σ1 and σ2 represent learning factors, r1 and r2 represent random numbers distributed in the interval [0, 1]. represents the individual optimal solution of the p-th particle at the o-th iteration, represents the position of the p-th particle at the o-th iteration, k is the k value of the kNN algorithm, is the jth particle in the neighboring solution set of the pth particle.

[0051] The particle position update formula is as follows:

[0052]

[0053] in, represents the position of the pth particle at the o+1th iteration.

[0054] Step D35: define a fitness threshold, repeat steps D32-D34 until the maximum number of iterations is reached or the optimal solution of the particle swarm reaches the fitness threshold, and use the optimal solution of the particle swarm as the optimal solution for the power generation of each photovoltaic device and the charging and discharging scheduling of the energy storage system.

[0055] The beneficial results achieved by the present invention using the above scheme are as follows:

[0056] (1) In view of the problem that the prediction model only focuses on the dependency of time series and it is difficult to fully extract the characteristic information of the data, resulting in inaccurate prediction results, the present invention designs a fusion model based on multiple data, combines weather characteristics and periodic characteristics, and uses a variety of deep learning algorithms to predict photovoltaic power generation, load power consumption and load demand. Specifically, the fusion model based on multiple data uses the LSTM model to deeply mine the long-term dependency in the time series data, captures the bidirectional context information features through the Bi-LSTM model, and finely extracts the local information features through the convolution module. By effectively fusing these three features and further performing deep learning and prediction, a high-precision prediction of the performance of the photovoltaic storage system is achieved;

[0057] (2) In order to solve the problem that when using the optimization algorithm to schedule the energy of the nodes of the photovoltaic storage system, the prediction results of photovoltaic power generation and load demand are not effectively combined, resulting in the scheduling scheme being unable to adapt to the actual photovoltaic power generation fluctuations and load changes, the present invention optimizes the charging and discharging scheduling of photovoltaic power generation and energy storage systems through an improved particle swarm optimization algorithm. By combining the prediction results of photovoltaic power generation and load demand while fully considering the possible errors in the prediction results, the economic cost minimization objective function and multiple constraints are designed, and the speed update formula of the particle swarm algorithm is improved in combination with the kNN algorithm. The economic cost of operation is significantly reduced while ensuring the supply and demand balance of the photovoltaic storage system. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A framework diagram of an intelligent management system for optical storage equipment proposed by the present invention;

[0059] Figure 2 This is a framework diagram of the optical storage management platform proposed by the present invention;

[0060] Figure 3 A flow chart of the performance prediction module proposed by the present invention;

[0061] Figure 4 This is a framework diagram of the multi-data based fusion model proposed in the present invention.

[0062] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0063] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0064] Example 1, see Figure 1 and Figure 2 , Figure 1 This is a framework diagram of an intelligent management system for optical storage equipment proposed by the present invention. Figure 2 This is a framework diagram of the photovoltaic storage management platform proposed in the present invention. In this embodiment, an intelligent management system for photovoltaic storage equipment is proposed, and the specific contents are as follows:

[0065] An intelligent management system for photovoltaic storage equipment, specifically including photovoltaic equipment, photovoltaic inverters, energy storage systems, loads, distribution networks, transformers and a photovoltaic storage management platform, wherein the photovoltaic storage management platform specifically includes a data acquisition module, a performance prediction module, an optimization scheduling module and a fault diagnosis module;

[0066] The data acquisition module obtains the output power time series data of each photovoltaic device, the input power time series data of the load and the local weather time series data. The output power time series data of each photovoltaic device and the input power time series data of the load are obtained by collecting the data of the photovoltaic inverter and the load respectively, and the weather time series data is obtained through the public data of the local meteorological station;

[0067] The performance prediction module builds two fusion models based on multiple data with the same structure to respectively perform photovoltaic power generation prediction and load power consumption prediction, and then calculate the load demand prediction;

[0068] The optimization scheduling module takes the power generation output power of photovoltaic equipment, the discharge output power of energy storage system and the charging output power of energy storage system as decision variables, defines the objective function and constraint conditions, combines the results of photovoltaic power generation forecast and load demand forecast, and uses the improved particle swarm algorithm to optimize the power generation of each photovoltaic equipment and the charging and discharging of the energy storage system.

[0069] The fault diagnosis module monitors various system parameters in real time, sets the normal working range, and automatically issues an alarm when the parameters exceed the normal working range;

[0070] The weather time series data in this embodiment specifically includes light intensity, solar radiation, temperature, precipitation and air pollution level.

[0071] Example 2, see Figure 3 and Figure 4 , Figure 3 This is a flow chart of the performance prediction module proposed by the present invention. Figure 4 This is a framework diagram of the multi-data fusion model proposed in the present invention. Based on the above embodiments, this embodiment implements a performance prediction module. The specific contents are as follows:

[0072] The performance prediction module constructs two fusion models based on multiple data with the same structure to perform photovoltaic power generation prediction and load power consumption prediction respectively. The process of calculating the load demand prediction includes the following steps:

[0073] Step S1: convert the output power time series data of each photovoltaic device into an embedding vector to obtain photovoltaic power generation embedding, and convert the input power time series data of the load into an embedding vector to obtain load power embedding;

[0074] Step S2: convert the light intensity, solar radiation, temperature, precipitation and air pollution level into embedded vectors in turn and concatenate them, and perform linear transformation on the concatenated results to obtain weather characteristics;

[0075] Step S3: extract the date part of the output power time series data of each photovoltaic device, retain the month and day of the date, convert the date into an embedded vector, and obtain the periodic feature;

[0076] Step S4: construct a first fusion model based on multiple data and a second fusion model based on multiple data, wherein the structures of the first fusion model based on multiple data and the second fusion model based on multiple data are consistent, and the first fusion model based on multiple data specifically includes an input layer, a first LSTM layer, a second LSTM layer, a Bi-LSTM layer, a first Conv1d layer, a second Conv1d layer, a maximum pooling layer, a first dense layer, and a second dense layer;

[0077] Step S5: embedding photovoltaic power generation, weather characteristics and periodic characteristics are spliced ​​and put into the first multi-data based fusion model to obtain the photovoltaic power generation prediction result of each photovoltaic device;

[0078] Step S6: embedding the load power consumption, weather characteristics and period characteristics, and splicing them into the second multi-data based fusion model to obtain the load power consumption prediction result;

[0079] Step S7: Add the photovoltaic power generation prediction results of each photovoltaic device and subtract the load power consumption prediction result to obtain the load demand prediction result.

[0080] The specific steps of constructing the first multi-data based fusion model in step S4 are as follows:

[0081] Step S41: taking the data of the input layer as an input sequence;

[0082] Step S42: inputting the input sequence into the first LSTM layer, and inputting the output result of the first LSTM layer into the second LSTM layer to obtain LSTM features; the first LSTM layer and the second LSTM layer each include an LSTM model and a Dropout layer connected in sequence in a unidirectional manner, the units parameter of the LSTM model is set to 512, tanh is used as the activation function, and the dropout rate of the Dropout layer is set to 0.2;

[0083] Step S43: Input the input sequence to the Bi-LSTM layer to obtain Bi-LSTM features. The Bi-LSTM layer includes a Bi-LSTM model and a Dropout layer which are connected unidirectionally in sequence. The units parameter of the Bi-LSTM model is set to 256, and the dropout rate of the Dropout layer is set to 0.4.

[0084] Step S44: Input the input sequence to the first Convld layer, input the output result of the first Conv1d layer to the second Conv1d layer, and input the output result of the second Convld layer to the maximum pooling layer to obtain convolutional features. The filters parameter of the first Convld layer is set to 3, the kernel_size parameter is set to 1, the filters parameter of the second Convld layer is set to 5, and the kernel_size parameter is set to 3. The specific formula is as follows:

[0085] The specific formula of the convolution operation of the Convld layer is as follows:

[0086]

[0087] Among them, yq,w represents the output of the convolution operation of the (q, w) coordinates, M and K represent the maximum values ​​of the horizontal and vertical coordinates respectively, and X q+m，w+n Data representing the (q+m, w+n) coordinates of the input data, F m，n Represents the value of the filter at the (m, n) coordinate.

[0088] Step S45: concatenate the LSTM features, the Bi-LSTM features and the convolution features to obtain fusion features;

[0089] Step S46: Input the fused features into the first dense layer, and input the output result of the first dense layer into the second dense layer to obtain the output result of the first multi-data based fusion model. In this embodiment, the units parameter of the first dense layer is set to 256, and the units parameter of the second dense layer is set to 1. The activation functions of the first dense layer and the second dense layer are both linear.

[0090] The multi-data based fusion model in this embodiment is developed using Python 3.7 language and Tensorflow2.11.0 deep learning development framework. The corresponding CUDA version number is 11.2. Two NVIDIA Tesla V100 graphics cards are used for training. Each graphics card has 16GB of video memory, the CPU model is Intel Xeon Platinum 8160T, and the operating system is Ubuntu20.04.

[0091] Embodiment 3: Based on the above embodiment, this embodiment implements an optimized scheduling module, and the specific contents are as follows:

[0092] The specific steps of optimizing the scheduling module are as follows:

[0093] Step D1: Define the objective function of minimizing economic cost. The specific formula is as follows:

[0094]

[0095] Where min C represents the minimum economic cost, T represents the maximum time period, N represents the number of photovoltaic devices, and P i，t represents the power output of the ith photovoltaic device at time t, Indicates P i，t The square of a i 、b i 、c i are coefficients of variation, u i，t is the economic cost coefficient.

[0096] In this embodiment, the performance prediction module is used to obtain the photovoltaic power generation prediction results and load demand prediction results of each photovoltaic device every hour in the next day, both in MW. The maximum period is defined as 4, and the objective function is to minimize the economic cost every 4 hours. i 、b i 、c i They are defined as 0.0012, 22.4892 and 4890 respectively, and the economic cost coefficient of each photovoltaic device is 1 in each time period.

[0097] Step D2: Define constraints. Constraints include supply and demand balance constraints, energy storage system charging and discharging constraints, and photovoltaic equipment power generation constraints. The specific formula is as follows:

[0098] The specific formula of the supply and demand balance constraint is as follows:

[0099]

[0100] in, represents the discharge output power of the energy storage system at time t, represents the charging output power of the energy storage system at time t, It represents the load demand forecast result at the tth moment.

[0101] Define the maximum limit of the discharge output power of the energy storage system and the maximum limit of the charging output power of the energy storage system. The specific formula of the charging and discharging constraint conditions of the energy storage system is as follows:

[0102]

[0103] in, Indicates the maximum limit of the discharge output power of the energy storage system, Indicates the maximum limit of the charging output power of the energy storage system.

[0104] The specific formula of the photovoltaic equipment power generation constraint is as follows:

[0105]

[0106] 0≤P i，t ≤P max ;

[0107] in, represents the photovoltaic power generation prediction result of the i-th photovoltaic device at time t, δ is the deviation coefficient, P max is the maximum output power of the photovoltaic device. In this embodiment, δ is defined as 0.1.

[0108] Step D3: Use the improved particle swarm algorithm to optimize the scheduling of power generation of each photovoltaic device and charging and discharging of the energy storage system. The specific steps are as follows:

[0109] Step D31: define the maximum number of iterations and initialize the particle swarm. In this embodiment, the maximum number of iterations is defined as 200, the number of particle swarms initialized is 50, the learning factors σ1 and σ2 are both set to 2, the minimum inertia weight is set to 0.4, and the maximum inertia weight is set to 0.9;

[0110] Step D32: taking the minimized economic cost function as the fitness function and calculating the fitness of each particle;

[0111] Step D33: Update the optimal solution of the particle swarm and each particle according to the fitness. The smaller the fitness, the better the particle.

[0112] Step D34: Use the kNN algorithm to find the neighboring solution set of each particle and update the velocity and position of the particle. In this embodiment, the k value of the kNN algorithm is defined as 10. The specific formula is as follows:

[0113] The particle velocity update formula is as follows:

[0114]

[0115] in, and denote the speed of the p-th particle at the o+1th and oth iterations, ω o represents the inertia weight of the oth iteration, σ1 and σ2 represent learning factors, r1 and r2 represent random numbers distributed in the interval [0, 1]. represents the individual optimal solution of the p-th particle at the o-th iteration, represents the position of the p-th particle at the o-th iteration, k is the k value of the kNN algorithm, is the jth particle in the neighboring solution set of the pth particle.

[0116] The particle position update formula is as follows:

[0117]

[0118] in, represents the position of the pth particle at the o+1th iteration.

[0119] Step D35: define a fitness threshold, repeat steps D32-D34 until the maximum number of iterations is reached or the optimal solution of the particle swarm reaches the fitness threshold, and use the optimal solution of the particle swarm as the optimal solution for the power generation of each photovoltaic device and the charging and discharging scheduling of the energy storage system.

Claims

1. An intelligent management system for optical storage equipment, characterized in that: An intelligent management system for photovoltaic storage equipment includes photovoltaic equipment, photovoltaic inverters, energy storage systems, loads, distribution networks, transformers and a photovoltaic storage management platform, wherein the photovoltaic storage management platform specifically includes a data acquisition module, a performance prediction module, an optimization scheduling module and a fault diagnosis module; The data acquisition module obtains the output power time series data of each photovoltaic device, the input power time series data of the load and the local weather time series data; The performance prediction module constructs two fusion models based on multiple data with consistent structures, respectively performs photovoltaic power generation prediction and load power consumption prediction, and then calculates the load demand prediction; The optimization scheduling module uses the power generation output power of the photovoltaic equipment, the discharge output power of the energy storage system and the charging output power of the energy storage system as decision variables, combines the results of photovoltaic power generation prediction and load demand prediction, defines the objective function and constraint conditions, and uses the improved particle swarm algorithm to optimize the power generation of each photovoltaic equipment and the charging and discharging of the energy storage system. The fault diagnosis module monitors various parameters of the system in real time, sets a normal working range, and automatically issues an alarm when the parameters exceed the normal working range.

2. The intelligent management system for optical storage devices according to claim 1, characterized in that: The performance prediction module constructs two fusion models based on multiple data with the same structure, respectively performs photovoltaic power generation prediction and load power consumption prediction, and then calculates the load demand prediction process, which includes the following steps: Step S1: convert the output power time series data of each photovoltaic device into an embedding vector to obtain photovoltaic power generation embedding, and convert the input power time series data of the load into an embedding vector to obtain load power embedding; Step S2: convert each type of data in the weather time series data into an embedded vector in turn and concatenate them, and perform linear transformation on the concatenated results to obtain weather features; Step S3: extract the date part of the output power time series data of each photovoltaic device, retain the month and day of the date, convert the date into an embedded vector, and obtain the periodic feature; Step S4: construct a first fusion model based on multiple data and a second fusion model based on multiple data, wherein the structures of the first fusion model based on multiple data and the second fusion model based on multiple data are consistent, and the first fusion model based on multiple data specifically includes an input layer, a first LSTM layer, a second LSTM layer, a Bi-LSTM layer, a first Conv1d layer, a second Conv1d layer, a maximum pooling layer, a first dense layer, and a second dense layer; Step S5: embedding photovoltaic power generation, weather characteristics and periodic characteristics are spliced ​​and put into the first multi-data based fusion model to obtain the photovoltaic power generation prediction result of each photovoltaic device; Step S6: embedding the load power consumption, weather characteristics and period characteristics, and splicing them into the second multi-data based fusion model to obtain the load power consumption prediction result; Step S7: Add the photovoltaic power generation prediction results of each photovoltaic device and subtract the load power consumption prediction result to obtain the load demand prediction result.

3. The intelligent management system for optical storage devices according to claim 2, characterized in that: The specific steps of constructing the first multi-data based fusion model in step S4 are as follows: Step S41: taking the data of the input layer as an input sequence; Step S42: inputting the input sequence into the first LSTM layer, and inputting the output result of the first LSTM layer into the second LSTM layer to obtain LSTM features; the first LSTM layer and the second LSTM layer each include an LSTM model and a Dropout layer connected in sequence in a unidirectional manner; Step S43: Input the input sequence to the Bi-LSTM layer to obtain Bi-LSTM features. The Bi-LSTM layer includes a Bi-LSTM model and a Dropout layer that are unidirectionally connected in sequence. Step S44: input the input sequence to the first Conv1d layer, input the output result of the first Conv1d layer to the second Conv1d layer, and input the output result of the second Conv1d layer to the maximum pooling layer to obtain convolution features; Step S45: concatenate the LSTM features, the Bi-LSTM features and the convolution features to obtain fusion features; Step S46: Input the fused features into the first dense layer, and input the output result of the first dense layer into the second dense layer to obtain the output result of the first multi-data based fusion model.

4. The intelligent management system for optical storage devices according to claim 1, characterized in that: The objective function is specifically to minimize the economic cost, and the specific formula is as follows: Where minC represents the minimization of economic cost, T represents the maximum time period, N represents the number of photovoltaic devices, and P i,t represents the power output of the ith photovoltaic device at time t, Indicates P i,t The square of a i 、b i 、c i are coefficients of variation, u i,t is the economic cost coefficient.

5. The intelligent management system for optical storage devices according to claim 2, characterized in that: The constraints specifically include supply and demand balance constraints, energy storage system charging and discharging constraints, and photovoltaic equipment power generation constraints; The specific formula of the supply and demand balance constraint condition is as follows: in, represents the discharge output power of the energy storage system at time t, represents the charging output power of the energy storage system at time t, represents the load demand forecast result at time t; The specific formula of the energy storage system charge and discharge constraint condition is as follows: in, Indicates the maximum limit of the discharge output power of the energy storage system, Indicates the maximum limit of the charging output power of the energy storage system; The specific formula of the photovoltaic equipment power generation constraint condition is as follows: 0≤P i,t ≤P max ; in, represents the photovoltaic power generation prediction result of the i-th photovoltaic device at time t, δ is the deviation coefficient, P max is the maximum output power of the photovoltaic device.

Images