Mobile solar photovoltaic power supply energy storage method and device

Abstract

The invention discloses a mobile solar photovoltaic power supply energy storage method and device. The mobile solar photovoltaic power supply energy storage method comprises the following steps: constructing a multi-dimensional real-time characteristic matrix according to power information, battery temperature information, load information, meteorological data and solar basic information; inputting the multi-dimensional real-time feature matrix into a trained scene classification model so as to obtain a current scene label; selecting a trained stable scene photovoltaic prediction sub-model or a trained complex scene photovoltaic prediction sub-model as a model for the current scene according to the current scene label, and inputting the multi-dimensional real-time feature matrix into the model for the current scene so as to obtain predicted photovoltaic power; and inputting the multi-dimensional real-time feature matrix into the trained load prediction model so as to obtain an output load prediction value. According to the method, load and photovoltaic power collaborative prediction provides a comprehensive basis for energy storage management, the energy utilization efficiency is improved, and the service life of equipment is prolonged.

Description

Technical Field

[0001] This application relates to the field of solar photovoltaic technology, specifically to a mobile solar photovoltaic power storage method and device. Background Technology

[0002] Currently, energy storage management of mobile solar photovoltaic (PV) power sources largely adopts the traditional "real-time acquisition-simple control" model. This involves the PV controller acquiring real-time data on PV panel output power, battery SOC (State of Charge), and load current, and then controlling charge and discharge based on fixed thresholds (e.g., charging when PV power > load power, discharging otherwise). While this model is simple in structure and low in cost, it has significant technical limitations under complex operating conditions. For example, solar energy resources are highly susceptible to meteorological conditions (such as cloud movement, fluctuations in light intensity, and temperature changes), and existing technologies generally use single prediction models (such as statistical models based on historical averages or simple linear regression models) for PV power prediction, failing to consider the dynamic differences in scenarios. The load types of mobile PV power sources are characterized by "diversity and volatility" (e.g., in field operations, exploration equipment is the primary focus during the day, switching to lighting and communication equipment at night; in emergency rescue, the load power dynamically changes with the intensity of the rescue mission). Existing technologies mostly employ "fixed load curves" or "real-time tracking" management methods, without establishing a prediction mechanism based on load type. Summary of the Invention

[0003] The purpose of this invention is to provide a mobile solar photovoltaic power storage method to at least solve one of the above-mentioned technical problems.

[0004] One aspect of the present invention provides a mobile solar photovoltaic power storage method, the mobile solar photovoltaic power storage method comprising: Acquire power information, battery temperature information, load type information, meteorological data, and basic solar energy information; A multi-dimensional real-time feature matrix is ​​constructed based on power information, battery temperature information, load information, meteorological data, and basic solar energy information. Obtain the trained scene classification model, the trained stable scene photovoltaic prediction sub-model, the trained complex scene photovoltaic prediction sub-model, and the trained load prediction model. The multi-dimensional real-time feature matrix is ​​input into the trained scene classification model to obtain the current scene label; Based on the current scene label, a trained stable scene photovoltaic prediction sub-model or a trained complex scene photovoltaic prediction sub-model is selected as the model for the current scene, and a multi-dimensional real-time feature matrix is ​​input into the model for the current scene to obtain the predicted photovoltaic power. The multi-dimensional real-time feature matrix is ​​input into the trained load prediction model to obtain the output load prediction value.

[0005] Optionally, the mobile solar photovoltaic power storage method further includes: The photovoltaic prediction sub-model for stable scenarios, the photovoltaic prediction sub-model for complex scenarios, and the load prediction model were trained respectively.

[0006] Optionally, training the stable scenario photovoltaic prediction sub-model includes: Acquire historical information; the historical information includes historical radiation intensity information, historical spectral distribution information, historical radiation duration information, historical surface temperature information, historical effective light-receiving area information, historical incident angle correction coefficient information, historical open-circuit voltage information, historical short-circuit current information, historical fill factor, historical conversion efficiency information, historical line resistance information, historical transmission loss rate information, historical voltage drop information, historical charging current information, historical initial state of charge (SOC) information, historical charge and discharge efficiency information, historical real-time power information, historical power fluctuation coefficient information, and historical critical load percentage information. Based on the energy transfer sequence, historical radiation intensity information, historical spectral distribution information, historical radiation duration information, historical surface temperature information, historical effective light-receiving area information, historical incident angle correction coefficient information, historical open-circuit voltage information, historical short-circuit current information, historical fill factor information, historical conversion efficiency information, historical line resistance information, historical transmission loss rate information, historical voltage drop information, historical charging current information, historical initial SOC information, historical charge and discharge efficiency information, historical real-time power information, historical power fluctuation coefficient information, and historical critical load proportion information are divided into multiple node features with a transfer sequence. These node features include historical solar radiation node features, historical photovoltaic panel receiving node features, historical photoelectric conversion node features, historical circuit transmission node features, historical energy storage buffer node features, and historical load consumption node features. Obtain the solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model that are combined in series to form a stable scenario photovoltaic prediction sub-model. The solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model are trained using historical solar radiation node features, historical photovoltaic receiving node features, historical photoelectric conversion node features, historical circuit transmission node features, historical energy storage buffer node features, and historical load consumption node features, respectively, to obtain a trained stable scenario photovoltaic prediction sub-model.

[0007] Optionally, the step of training the solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model using historical solar radiation node features, historical photovoltaic receiving node features, historical photoelectric conversion node features, historical circuit transmission node features, historical energy storage buffer node features, and historical load consumption node features, respectively, to obtain the trained stable scenario photovoltaic prediction sub-model, includes: The historical radiation intensity information, historical spectral distribution information, historical radiation persistence information, and actual atmospheric attenuation loss values ​​are preprocessed to obtain the preprocessed data for training the solar radiation node sub-model. Obtain a solar radiation node sub-model, wherein the solar radiation node sub-model is a multiple linear regression model; The preprocessed training solar radiation node sub-model is used to train the solar radiation node sub-model with data.

[0008] Optionally, the step of training the solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model using historical solar radiation node features, historical photovoltaic panel receiving node features, historical photovoltaic conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model, respectively, to obtain the trained stable scenario photovoltaic prediction sub-model, further includes: The historical surface temperature information, historical effective light-receiving area information, and historical incident angle correction coefficient information are preprocessed to obtain the preprocessed data for training the photovoltaic panel receiver node sub-model. Obtain the effective energy data of the solar radiation nodes output by the solar radiation node sub-model during the training process; Obtain a photovoltaic panel receiving node sub-model, wherein the photovoltaic panel receiving node sub-model is a linear regression model; The photovoltaic panel receiving node sub-model is trained using preprocessed training data and effective energy data of solar radiation nodes.

[0009] Optionally, the step of training the solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model using historical solar radiation node features, historical photovoltaic panel receiving node features, historical photovoltaic conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model, respectively, to obtain the trained stable scenario photovoltaic prediction sub-model, further includes: The historical photoelectric conversion node feature data is preprocessed to obtain the preprocessed data for training the photoelectric conversion node sub-model. Obtain the output energy of the historical photovoltaic receiving node features output by the photovoltaic receiving node sub-model obtained during the training of the photovoltaic receiving node sub-model; Obtain a photoelectric conversion node sub-model, wherein the photoelectric conversion node sub-model is a multiple linear regression model; The photoelectric conversion node sub-model is trained using preprocessed training data and the output energy of historical photovoltaic panel receiving node characteristics.

[0010] Optionally, the step of training the solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model using historical solar radiation node features, historical photovoltaic panel receiving node features, historical photovoltaic conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model, respectively, to obtain the trained stable scenario photovoltaic prediction sub-model, further includes: The historical circuit transmission node feature data is preprocessed to obtain the preprocessed data for training the circuit transmission node sub-model. Obtain the output energy of the historical photoelectric conversion node features output by the photoelectric conversion node sub-model obtained during the training process; Obtain a circuit transmission node sub-model, wherein the circuit transmission node sub-model is a multiple linear regression model; The circuit transmission node sub-model is trained using preprocessed training circuit transmission node data and the output energy of historical photoelectric conversion node features.

[0011] Optionally, the step of training the solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model using historical solar radiation node features, historical photovoltaic panel receiving node features, historical photovoltaic conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model, respectively, to obtain the trained stable scenario photovoltaic prediction sub-model, further includes: The historical energy storage buffer node feature data is preprocessed to obtain the preprocessed data for training the energy storage buffer node sub-model. Obtain the output energy of the historical circuit transmission node features output by the circuit transmission node sub-model obtained during the training of the circuit transmission node sub-model; Obtain the energy storage buffer node sub-model, wherein the energy storage buffer node sub-model is a multiple linear regression model; The energy storage buffer node sub-model is trained using preprocessed training data and the output energy of historical photoelectric conversion node characteristics.

[0012] Optionally, the step of training the solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model using historical solar radiation node features, historical photovoltaic panel receiving node features, historical photovoltaic conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model, respectively, to obtain the trained stable scenario photovoltaic prediction sub-model, further includes: The historical load consumption node feature data is preprocessed to obtain the preprocessed data for training the load consumption node sub-model. Obtain the output energy of the historical energy storage buffer node features output by the energy storage buffer node sub-model obtained during the training of the energy storage buffer node sub-model; Obtain the load consumption node sub-model, wherein the load consumption node sub-model is a multiple linear regression model; The load consumption node sub-model is trained using preprocessed training data and the output energy of historical energy storage buffer node features.

[0013] This application also provides a mobile solar photovoltaic power storage device, the mobile solar photovoltaic power storage device comprising: The basic information acquisition module is used to acquire power information, battery temperature information, load type information, meteorological data, and basic solar energy information. The feature matrix acquisition module is used to construct a multi-dimensional real-time feature matrix based on power information, battery temperature information, load information, meteorological data, and basic solar energy information. The model acquisition module is used to acquire trained scene classification models, trained stable scene photovoltaic prediction sub-models, trained complex scene photovoltaic prediction sub-models, and trained load prediction models. A scene label acquisition module is used to input the multi-dimensional real-time feature matrix into a trained scene classification model to obtain the current scene label. A photovoltaic power prediction module is used to select a trained stable scene photovoltaic prediction sub-model or a trained complex scene photovoltaic prediction sub-model as the model for the current scene based on the current scene label, and input a multi-dimensional real-time feature matrix into the model for the current scene to obtain the predicted photovoltaic power. The load prediction module is used to input the multi-dimensional real-time feature matrix into the trained load prediction model to obtain the output load prediction value.

[0014] This application's mobile solar photovoltaic power storage method constructs a real-time feature matrix by collecting multi-dimensional information, achieving comprehensive perception of the system's operating status and laying a data foundation for accurate prediction. This application employs a scenario classification and sub-model adaptation architecture, adaptively selecting the prediction model based on real-time features to solve the problem of poor adaptability of a single model and improve prediction accuracy for different scenarios. In stable scenarios, a series-connected node sub-model design accurately depicts the entire energy transfer process. Combined with targeted training and data reuse, it balances prediction accuracy and computational efficiency, meeting the real-time requirements of mobile scenarios. Coordinated prediction of load and photovoltaic power provides comprehensive data for energy storage management, facilitating the optimization of charging and discharging strategies and improving energy utilization efficiency and equipment lifespan. Attached Figure Description

[0015] Figure 1 This is a schematic flowchart of a mobile solar photovoltaic power storage method according to an embodiment of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are some, but not all, embodiments of this application. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0017] like Figure 1 The mobile solar photovoltaic power storage method shown includes: Acquire power information, battery temperature information, load type information, meteorological data, and basic solar energy information; A multi-dimensional real-time feature matrix is ​​constructed based on power information, battery temperature information, load information, meteorological data, and basic solar energy information. Obtain the trained scene classification model, the trained stable scene photovoltaic prediction sub-model, the trained complex scene photovoltaic prediction sub-model, and the trained load prediction model. The multi-dimensional real-time feature matrix is ​​input into the trained scene classification model to obtain the current scene label; Based on the current scene label, a trained stable scene photovoltaic prediction sub-model or a trained complex scene photovoltaic prediction sub-model is selected as the model for the current scene, and a multi-dimensional real-time feature matrix is ​​input into the model for the current scene to obtain the predicted photovoltaic power. The multi-dimensional real-time feature matrix is ​​input into the trained load prediction model to obtain the output load prediction value.

[0018] In this embodiment, the mobile solar photovoltaic power storage method further includes: The photovoltaic prediction sub-model for stable scenarios, the photovoltaic prediction sub-model for complex scenarios, and the load prediction model were trained respectively.

[0019] In this embodiment, training the stable scenario photovoltaic prediction sub-model includes: Acquire historical information; the historical information includes historical radiation intensity information, historical spectral distribution information, historical radiation duration information, historical surface temperature information, historical effective light-receiving area information, historical incident angle correction coefficient information, historical open-circuit voltage information, historical short-circuit current information, historical fill factor, historical conversion efficiency information, historical line resistance information, historical transmission loss rate information, historical voltage drop information, historical charging current information, historical initial state of charge (SOC) information, historical charge and discharge efficiency information, historical real-time power information, historical power fluctuation coefficient information, and historical critical load percentage information. Based on the energy transfer sequence, historical radiation intensity information, historical spectral distribution information, historical radiation duration information, historical surface temperature information, historical effective light-receiving area information, historical incident angle correction coefficient information, historical open-circuit voltage information, historical short-circuit current information, historical fill factor information, historical conversion efficiency information, historical line resistance information, historical transmission loss rate information, historical voltage drop information, historical charging current information, historical initial SOC information, historical charge and discharge efficiency information, historical real-time power information, historical power fluctuation coefficient information, and historical critical load proportion information are divided into multiple node features with a transfer sequence. These node features include historical solar radiation node features, historical photovoltaic panel receiving node features, historical photoelectric conversion node features, historical circuit transmission node features, historical energy storage buffer node features, and historical load consumption node features. Obtain the solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model that are combined in series to form a stable scenario photovoltaic prediction sub-model. The solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model are trained using historical solar radiation node features, historical photovoltaic receiving node features, historical photoelectric conversion node features, historical circuit transmission node features, historical energy storage buffer node features, and historical load consumption node features, respectively, to obtain a trained stable scenario photovoltaic prediction sub-model.

[0020] In this embodiment, the process of training the solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model using historical solar radiation node features, historical photovoltaic receiving node features, historical photoelectric conversion node features, historical circuit transmission node features, historical energy storage buffer node features, and historical load consumption node features, respectively, to obtain the trained stable scenario photovoltaic prediction sub-model, includes: The historical radiation intensity information, historical spectral distribution information, historical radiation persistence information, and actual atmospheric attenuation loss values ​​are preprocessed to obtain the preprocessed data for training the solar radiation node sub-model. Obtain a solar radiation node sub-model, wherein the solar radiation node sub-model is a multiple linear regression model; The preprocessed training solar radiation node sub-model is used to train the solar radiation node sub-model with data.

[0021] In this embodiment, the model expression for the solar radiation node sub-model is as follows: ;in, It is the natural logarithm. For a local minimum (e.g., 0.01), exp is an exponential function. V is an absolute value function, reflecting the deviation of the visible light proportion from the median value; k16-k46 are characteristic coefficients, and b is a bias term.

[0022] In this embodiment, the above model expression is used, and through logarithmic and exponential processing, the nonlinear physical process of atmospheric decay is accurately characterized while maintaining the core of "linear combination of coefficients".

[0023] In this embodiment, historical solar radiation intensity data is obtained from a radiation sensor, covering multiple full years, and sampled at fixed time intervals.

[0024] In this embodiment, historical spectral distribution information (such as the proportion of ultraviolet, visible, and infrared light) is used to reflect the spectral composition of solar radiation and is synchronized with the radiation intensity data over time.

[0025] In this embodiment, historical radiation persistence information describes the stability of radiation intensity, such as fluctuations over multiple consecutive sampling periods.

[0026] In this embodiment, preprocessing may include missing value handling, outlier removal, and data alignment.

[0027] In this embodiment, the step of training the solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model using historical solar radiation node features, historical photovoltaic receiving node features, historical photoelectric conversion node sub-model, historical circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model, respectively, to obtain the trained stable scenario photovoltaic prediction sub-model, further includes: The historical surface temperature information, historical effective light-receiving area information, and historical incident angle correction coefficient information are preprocessed to obtain the preprocessed data for training the photovoltaic panel receiver node sub-model. Obtain the effective energy data of the solar radiation nodes output by the solar radiation node sub-model during the training process; Obtain a photovoltaic panel receiving node sub-model, wherein the photovoltaic panel receiving node sub-model is a linear regression model; The photovoltaic panel receiving node sub-model is trained using preprocessed training data and effective energy data of solar radiation nodes.

[0028] In this embodiment, the time span of the historical surface temperature information is consistent with the training data of the solar radiation node model, and the sampling frequency is 15 minutes / time; In this embodiment, the historical effective light-receiving area information comes from the characteristics of the photovoltaic panel receiving node and reflects the actual area of ​​radiation received.

[0029] In this embodiment, the historical incident angle correction coefficient information is used to correct the impact of the incident angle on the received energy.

[0030] In this embodiment, the data used to train the photovoltaic panel receiver node sub-model also includes the actual value of temperature-induced additional losses ( ), through formula =Effective energy data of solar radiation node E1 × Effective light-receiving area × Incident angle correction factor × ( Calculate using ) / 100, where The conversion efficiency benchmark value This represents the conversion efficiency at the actual temperature.

[0031] In this embodiment, preprocessing may include missing value handling, outlier removal, and data alignment.

[0032] In this embodiment, the model expression for the photovoltaic panel receiving node sub-model is: Where k1 is the temperature coefficient and b is the bias term.

[0033] In this embodiment, the photovoltaic panel receiver node sub-model uses mean square error (MSE) as the optimization objective, and the formula is as follows: , where n is the number of samples.

[0034] In this embodiment, the step of training the solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model using historical solar radiation node features, historical photovoltaic receiving node features, historical photoelectric conversion node sub-model, historical circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model, respectively, to obtain the trained stable scenario photovoltaic prediction sub-model, further includes: The historical photoelectric conversion node feature data is preprocessed to obtain the preprocessed data for training the photoelectric conversion node sub-model. Obtain the output energy of the historical photovoltaic receiving node features output by the photovoltaic receiving node sub-model obtained during the training of the photovoltaic receiving node sub-model; Obtain a photoelectric conversion node sub-model, wherein the photoelectric conversion node sub-model is a multiple linear regression model; The photoelectric conversion node sub-model is trained using preprocessed training data and the output energy of historical photovoltaic panel receiving node characteristics.

[0035] In this embodiment, the historical photoelectric conversion node characteristic data includes historical open-circuit voltage information, used to reflect the voltage characteristics of the photovoltaic panel under no-load conditions; historical short-circuit current information, used to reflect the current characteristics of the photovoltaic panel under short-circuit conditions; historical fill factor information, used to reflect the performance of the photovoltaic panel at its maximum power point; historical conversion efficiency information, used to characterize the energy conversion ratio of light to electricity; and actual conversion loss value (…). ): Through formula =The output energy E2 of the historical photovoltaic panel receiving node characteristics is calculated as (open circuit voltage × short circuit current × fill factor).

[0036] In this embodiment, preprocessing may include missing value handling, outlier removal, and data alignment.

[0037] In this embodiment, the model expression of the photoelectric conversion node sub-model is as follows: Where k1, k2, and k3 are characteristic coefficients, and b is a bias term; k15 is used to capture the basic loss ratio of E2, k25 is used to reflect the impact of efficiency deviation on loss, and k35 is used to reflect the additional loss caused by insufficient fill factor. In this embodiment, the loss function of the photoelectric conversion node sub-model is: weight Strengthen the high-loss ratio sample ( The fitting accuracy of ).

[0038] In this embodiment, the step of training the solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model using historical solar radiation node features, historical photovoltaic receiving node features, historical photoelectric conversion node sub-model, historical circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model, respectively, to obtain the trained stable scenario photovoltaic prediction sub-model, further includes: The historical circuit transmission node feature data is preprocessed to obtain the preprocessed data for training the circuit transmission node sub-model. Obtain the output energy of the historical photoelectric conversion node features output by the photoelectric conversion node sub-model obtained during the training process; Obtain a circuit transmission node sub-model, wherein the circuit transmission node sub-model is a multiple linear regression model; The circuit transmission node sub-model is trained using preprocessed training circuit transmission node data and the output energy of historical photoelectric conversion node features.

[0039] In this embodiment, the historical circuit transmission node characteristic data includes historical line resistance information, used to reflect the resistance characteristics of the transmission line; historical transmission loss rate information, used to characterize the proportion of energy loss during transmission; historical voltage drop information, used to reflect the voltage drop when current passes through the line; and the actual value of transmission loss, which can be expressed by formula L4. act =The output energy E3 of the historical photoelectric conversion node characteristics is calculated by multiplying the transmission loss rate by 100. The actual loss is derived based on the loss rate data of the circuit transmission node characteristics.

[0040] In this embodiment, preprocessing may include missing value handling, outlier removal, and data alignment.

[0041] In this embodiment, the model expression for the photoelectric conversion node sub-model is: Among them, k21, k22, k23, and k24 are characteristic coefficients, b is a bias term, k21 is used to capture the basic transmission loss ratio of E3, k22 is used to reflect the loss caused by line resistance, and k23 is used to reflect the impact of loss rate deviation on total loss. In this embodiment, the loss function of the photoelectric conversion node sub-model adopts the weighted mean square error (WMSE), with weights... Strengthen the high-loss ratio sample ( The fitting accuracy of ).

[0042] In this embodiment, the step of training the solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model using historical solar radiation node features, historical photovoltaic receiving node features, historical photoelectric conversion node sub-model, historical circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model, respectively, to obtain the trained stable scenario photovoltaic prediction sub-model, further includes: The historical energy storage buffer node feature data is preprocessed to obtain the preprocessed data for training the energy storage buffer node sub-model. Obtain the output energy of the historical circuit transmission node features output by the circuit transmission node sub-model obtained during the training of the circuit transmission node sub-model; Obtain the energy storage buffer node sub-model, wherein the energy storage buffer node sub-model is a multiple linear regression model; The energy storage buffer node sub-model is trained using preprocessed training data and the output energy of historical photoelectric conversion node characteristics.

[0043] In this embodiment, the historical energy storage buffer node characteristic data includes historical charging current information, used to reflect the magnitude of the charging current of the energy storage device; historical initial state of charge (SOC) information, used to characterize the initial state of charge of the energy storage device; historical charge and discharge efficiency information, used to reflect the energy conversion efficiency during the energy storage process; and the actual value of energy storage loss, expressed by the formula... =The output energy of the historical circuit transmission node characteristics is calculated as E4 × (1 - charge / discharge efficiency / 100).

[0044] In this embodiment, preprocessing may include missing value handling, outlier removal, and data alignment.

[0045] In this embodiment, the model expression for the load consumption node sub-model is: Among them, k11, k12, k13, and k14 are characteristic coefficients, b is a bias term, k11 is used to capture the basic energy storage loss ratio of E4, k12 is used to reflect the impact of charging load on loss, k13 is used to reflect the additional loss when the initial value of SOC is low, and k14 reflects the impact of efficiency deviation. In this embodiment, the loss function of the load consumption node sub-model is the weighted mean squared error (WMSE), with weights... Strengthen the high-loss ratio sample ( The fitting accuracy of ).

[0046] In this embodiment, the step of training the solar radiation node sub-model, photovoltaic panel receiving node sub-model, photoelectric conversion node sub-model, circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model using historical solar radiation node features, historical photovoltaic receiving node features, historical photoelectric conversion node sub-model, historical circuit transmission node sub-model, energy storage buffer node sub-model, and load consumption node sub-model, respectively, to obtain the trained stable scenario photovoltaic prediction sub-model, further includes: The historical load consumption node feature data is preprocessed to obtain the preprocessed data for training the load consumption node sub-model. Obtain the output energy of the historical energy storage buffer node features output by the energy storage buffer node sub-model obtained during the training of the energy storage buffer node sub-model; Obtain the load consumption node sub-model, wherein the load consumption node sub-model is a multiple linear regression model; The load consumption node sub-model is trained using preprocessed training data and the output energy of historical energy storage buffer node features.

[0047] In this embodiment, the historical load consumption node characteristic data may include historical real-time power information, used to reflect the current actual power consumption of the load; historical power fluctuation coefficient information, used to reflect the degree of load power fluctuation over a certain period of time; historical critical load proportion information, used to characterize the proportion of critical load power consumption to total load power; and the actual value of load matching loss L6. act In this embodiment, L6 act The output energy E5-real-time power can be obtained through the historical energy storage buffer node characteristics.

[0048] In this embodiment, preprocessing may include missing value handling, outlier removal, and data alignment.

[0049] In this embodiment, the model expression for the load consumption node sub-model is: ;in, Given the basic supply-demand deviation (i.e., the absolute value of (E5 - real-time power)), k1, k2, and k3 are characteristic coefficients, and b is the bias term; where k1 represents capturing the loss caused by the basic supply-demand deviation, k2 reflects the amplification effect of the fluctuation impact factor on the loss, and k3 reflects the suppression effect of the critical load ratio on the loss.

[0050] In this embodiment, the loss function of the load consumption node sub-model is the weighted mean squared error (WMSE), with weights... Strengthen the high-loss ratio sample ( The fitting accuracy of ).

[0051] In this embodiment, the complex scenario photovoltaic prediction sub-model includes a dynamic solar radiation sub-model, a dynamic photovoltaic panel receiving sub-model, a dynamic photoelectric conversion sub-model, a dynamic circuit transmission sub-model, a dynamic energy storage buffer sub-model, a dynamic load consumption sub-model, a Transformer fusion layer, and a physical constraint layer. In this embodiment, the expression for the dynamic solar radiation sub-model is as follows: ; in, Indicates the future number Predicted radiation intensity values ​​per minute; Indicates the speed of cloud movement (extracted from real-time satellite cloud imagery). The cloud cover change rate over 5 minutes.

[0052] For physical constraint terms ( (where k is the current cloud cover and k is the attenuation coefficient) to ensure that the radiation intensity does not violate the atmospheric attenuation law. The physical constraint weights (ranging from 0.3 to 0.5, balancing data fit with physical plausibility).

[0053] In this embodiment, the expression for the dynamic photovoltaic panel receiver sub-model is as follows: ;in, The output radiation energy of the dynamic solar radiation sub-model; Effective light-receiving area; This represents the proportion of local shadows. The shadow effect coefficient (values ​​range from 1.2 to 1.5, as shadows can cause hot spot effects on photovoltaic panels, resulting in losses exceeding the proportion of shadowed area). β represents the real-time pitch of the photovoltaic panel; β represents the real-time roll angle of the photovoltaic panel.

[0054] In this embodiment, the expression for the dynamic solar radiation sub-model is as follows: ; ; ; in, Indicates dynamic photoelectric conversion efficiency (after correction). This is the baseline value for conversion efficiency under standard conditions (e.g., 18%). This is the temperature coefficient (typical value 0.004°C). (Difference between photovoltaic panel temperature and standard temperature); This is the output of the dynamic photoelectric conversion sub-model; The threshold for the rate of temperature change (e.g., 2°C / min). This is the sudden change penalty coefficient (typical value 0.1); The fill factor (reflects the maximum power point performance of the photovoltaic panel, with a value of 0.7~0.85); Conversion loss (basic loss 5% + additional loss due to insufficient fill factor); This is a rectification function (outputs x when x>0, otherwise outputs 0), used to capture the nonlinear effects of sudden temperature changes. β represents the real-time pitch of the photovoltaic panel; β represents the real-time roll angle of the photovoltaic panel.

[0055] In this embodiment, the output of the dynamic circuit transmission sub-model is ; In this embodiment, the contact resistance fluctuation caused by vibration is predicted using a GRU. Combined with line resistance Calculate real-time transmission loss and output the energy after stable transmission.

[0056] In this embodiment, the formula for the dynamic circuit transmission sub-model is as follows: ; For the future Predicted contact resistance at any given time; For gated loop unit; This is the actual value of the contact resistance at the current time (time t); The contact resistance historical values ​​are from the previous 1 to 4 minutes. The vibration frequency at the current moment.

[0057] In this embodiment, Obtain it using the following formula: ; The current moment is the tilt angle of the photovoltaic panel i minutes ago; The elevation angle of the photovoltaic panel is i+1 minutes before the current moment; For time step.

[0058] In this embodiment, ;in, This is the output of the dynamic photoelectric conversion sub-model; In this embodiment, ;in, For the future Energy loss during transmission at any given moment; future The square of the current transmitted at any given moment; For the future Total transmission resistance at any given time; For time step; In this embodiment, ;in, For the future Total transmission resistance at any given time; For fixed resistance in transmission lines; For the future Predicted contact resistance value at any given time.

[0059] In this embodiment, ;in, For the future Energy loss during transmission at any given moment; For the future The square of the current transmitted at any given moment; For the future Total transmission resistance at any given time; For time step; In this embodiment, the expression for the dynamic energy storage buffer sub-model is as follows: ; in, For the future Dynamic charge and discharge efficiency at any given time; This represents the actual charging and discharging efficiency at the current moment. This represents the rate of change of the charging and discharging current at the current moment. This represents the instantaneous fluctuation value of SOC (State of Charge) at the current moment.

[0060] In this embodiment, the energy storage loss is calculated as follows: ; in, For the future Energy loss from energy storage at any given moment; The output of the dynamic circuit transmission sub-model; For the future Dynamic charge and discharge efficiency at any given time; This represents the instantaneous fluctuation value of SOC (State of Charge) at the current moment.

[0061] In this embodiment, ;in, Output for the dynamic energy storage buffer sub-model; The output of the dynamic circuit transmission sub-model; For the future Energy loss from energy storage at any given moment; In this embodiment, the expression for the dynamic load consumption sub-model is as follows: ;in, This represents the attention weight of the k-th type of load at time t; This represents the "importance score" of the k-th type of load at the current moment; In this embodiment, the "importance score" of the k-th type of load at the current time can be obtained by the following formula: ; in, The type embedding vector for the k-th load (dimension: This is obtained by encoding the load power characteristics (average power, maximum power); This is the current load power time series (including the load power history of the previous 5 minutes). In this embodiment, the dynamic load consumption sub-model is attention-LSTM.

[0062] In this embodiment, the various sub-models of the complex scenario photovoltaic prediction sub-model are connected in series, as described in detail below: Step 1: Dynamic Solar Radiation Sub-model (Energy Source) First, predict the future time-series solar radiation intensity. Outputs "initial radiation energy" This is the "energy source" for the entire photovoltaic power generation, providing the basic input for all subsequent nodes.

[0063] Step 2: Dynamic Photovoltaic Panel Receiving Sub-model (Energy Reception) The output of the first step As the core input, combined with attitude angles Shadow proportion Calculate the "effective received energy" using dynamic factors. This refers to the energy that a photovoltaic panel can actually capture, which directly connects the physical process of "radiated energy → received energy".

[0064] Step 3: Dynamic photoelectric conversion sub-model (energy conversion) Enter the second step Through the rate of temperature change Attitude correction coefficient Quantify conversion efficiency loss and output "converted electrical energy" ( This corresponds to the photoelectric conversion physical process of "receiving energy → electrical energy".

[0065] Step 4: Dynamic circuit transmission sub-model (energy transfer); With the third step As input, and taking into account factors such as vibration frequency and contact resistance fluctuations, the line transmission loss is calculated, and the output is "transmitted electrical energy" (…). It is the circuit loss link that connects "converting electrical energy to transmitting electrical energy".

[0066] Step 5: Dynamic Energy Storage Buffer Sub-model (Energy Buffer) Enter the fourth step Taking into account the rate of change of charging and discharging current and instantaneous fluctuations of SOC, the output is "stored acceptable electrical energy" ( — This refers to the actual electrical energy that the energy storage system can receive, corresponding to the energy regulation stage of "transmitting electrical energy → storing energy buffer".

[0067] Step 6: Dynamic Load Consumption Sub-model (Energy Feedback) With the fifth step As input, combined with load power surges and critical load switching signals, the output is "final available photovoltaic power" (…). — This refers to the net power available for energy storage after deducting the immediate consumption by the load, forming an energy feedback loop of "energy storage → load".

[0068] In this embodiment, the load prediction model adopts an architecture of "feature embedding layer + multi-branch GRU layer + correlation constraint layer + burst adjustment layer" to output the "partial load power sequence" and "total load power sequence" every 5 minutes within the next 30 minutes.

[0069] In this embodiment, the scene classification model adopts an architecture of "dynamic feature extraction layer + CNN-LSTM temporal layer + uncertainty quantization layer + classification output layer". It takes "multi-dimensional feature sequence of the past 10 minutes" as input and outputs the current scene label (stable / complex) and classification confidence.

[0070] The application also provides a mobile solar photovoltaic power storage device, the mobile solar photovoltaic power storage device comprising: The basic information acquisition module is used to acquire power information, battery temperature information, load type information, meteorological data, and basic solar energy information. The feature matrix acquisition module is used to construct a multi-dimensional real-time feature matrix based on power information, battery temperature information, load information, meteorological data, and basic solar energy information. The model acquisition module is used to acquire trained scene classification models, trained stable scene photovoltaic prediction sub-models, trained complex scene photovoltaic prediction sub-models, and trained load prediction models. A scene label acquisition module is used to input the multi-dimensional real-time feature matrix into a trained scene classification model to obtain the current scene label. A photovoltaic power prediction module is used to select a trained stable scene photovoltaic prediction sub-model or a trained complex scene photovoltaic prediction sub-model as the model for the current scene based on the current scene label, and input a multi-dimensional real-time feature matrix into the model for the current scene to obtain the predicted photovoltaic power. The load prediction module is used to input the multi-dimensional real-time feature matrix into the trained load prediction model to obtain the output load prediction value.

[0071] This application has the following advantages: By constructing a multi-dimensional real-time feature matrix and utilizing a trained scene classification model, the current scene label (stable / complex) is accurately determined. Based on different scene labels, appropriate stable scene photovoltaic prediction sub-models or complex scene photovoltaic prediction sub-models are selected for photovoltaic power prediction. This approach of using different prediction models for different scenes, compared to single-model prediction, can more accurately consider various complex factors affecting photovoltaic power under different scenarios, greatly improving the accuracy of photovoltaic power prediction and providing a reliable data foundation for subsequent energy storage and power distribution.

[0072] The stable-scenario photovoltaic prediction sub-model consists of multiple cascaded sub-models, each trained to address different node characteristics of energy transfer. Taking the solar radiation node sub-model as an example, it employs a multiple linear regression model and, through a unique model expression, accurately characterizes the nonlinear physical process of atmospheric attenuation while maintaining the core principle of "linear combination of coefficients." It fully considers the impact of historical radiation intensity information, historical spectral distribution information, and historical radiation persistence information on solar radiation, thereby improving the accuracy of solar radiation energy prediction.

[0073] From historical solar radiation node characteristics to historical load consumption node characteristics, each node characteristic is preprocessed and trained separately, fully exploring the factors affecting energy transfer and power changes at different stages. For example, the photovoltaic panel receiving node sub-model considers historical surface temperature information, historical effective light-receiving area information, and historical incident angle correction coefficient information. Through accurate model training, the energy received by the photovoltaic panel and related losses can be calculated more accurately, making the photovoltaic prediction sub-model for the entire stable scenario more comprehensive and accurate in predicting photovoltaic power.

[0074] The complex scenario photovoltaic prediction sub-model comprises multiple dynamic sub-models, combined with a Transformer fusion layer and a physical constraint layer. Taking the dynamic solar radiation sub-model as an example, its expression considers dynamic factors such as cloud movement speed and cloud cover change rate, while introducing physical constraint terms to ensure that radiation intensity does not violate atmospheric attenuation laws. By appropriately setting the weights of physical constraints, the model balances data fitting with physical rationality. This combined approach better addresses the rapid changes and uncertainties in solar radiation under complex scenarios, improving the accuracy of solar radiation energy prediction in complex environments.

[0075] The various dynamic sub-models are connected in series according to the energy transfer sequence. Starting with the initial radiant energy output from the dynamic solar radiation sub-model, they sequentially pass through the dynamic photovoltaic panel receiving sub-model, the dynamic photoelectric conversion sub-model, the dynamic circuit transmission sub-model, the dynamic energy storage buffer sub-model, and the dynamic load consumption sub-model. Each sub-model considers the dynamic factors of its corresponding stage. For example, the dynamic photovoltaic panel receiving sub-model considers the local shading ratio, the real-time pitch and roll angle of the photovoltaic panel, etc., ultimately forming an energy feedback closed loop from energy storage to load. This comprehensive and dynamic simulation method can more realistically reflect the energy conversion and transfer process of photovoltaic systems under complex scenarios, thereby more accurately predicting photovoltaic power under complex scenarios.

[0076] The entire mobile solar photovoltaic (PV) power storage method enables optimized control of the energy storage system through accurate prediction of PV power and load power. Based on the predicted PV power and load power, the charging and discharging strategies of the energy storage equipment are rationally arranged, improving the utilization efficiency of PV energy and reducing energy waste. Simultaneously, the various models work collaboratively, forming an organic whole from basic information acquisition to final power prediction and energy storage control. This enhances the stability, reliability, and operational efficiency of the entire mobile solar PV power storage system, allowing it to better adapt to different working scenarios and user needs.

[0077] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A mobile solar photovoltaic power source energy storage method, characterized in that, The mobile solar photovoltaic power supply energy storage method comprises the following steps: acquiring power information, battery temperature information, load type information, meteorological data, and solar basic information; constructing a multi-dimensional real-time feature matrix according to the power information, the battery temperature information, the load information, the meteorological data, and the solar basic information; acquiring a trained scene classification model, a trained stable scene photovoltaic prediction sub-model, a trained complex scene photovoltaic prediction sub-model, and a trained load prediction model; inputting the multi-dimensional real-time feature matrix into the trained scene classification model to acquire a current scene label; selecting the trained stable scene photovoltaic prediction sub-model or the trained complex scene photovoltaic prediction sub-model as a current scene model according to the current scene label, and inputting the multi-dimensional real-time feature matrix into the current scene model to acquire a predicted photovoltaic power; inputting the multi-dimensional real-time feature matrix into the trained load prediction model to acquire an output load prediction value.

2. The mobile solar photovoltaic power source energy storage method of claim 1, wherein, The mobile solar photovoltaic power supply energy storage method further comprises the following steps: training the stable scene photovoltaic prediction sub-model, the complex scene photovoltaic prediction sub-model, and the load prediction model respectively.

3. The mobile solar photovoltaic power source energy storage method of claim 2, wherein, The training of the stable scene photovoltaic prediction sub-model comprises the following steps: acquiring historical information; the historical information comprises historical radiation intensity information, historical spectral distribution information, historical radiation persistence information, historical surface temperature information, historical effective light-receiving area information, historical incident angle correction coefficient information, historical open-circuit voltage information, historical short-circuit current information, historical fill factor, historical conversion efficiency information, historical line resistance information, historical transmission loss rate information, historical voltage drop information, historical charging current information, historical SOC initial value information, historical charging and discharging efficiency information, historical real-time power information, historical power fluctuation coefficient information, and historical key load proportion information; dividing the historical radiation intensity information, the historical spectral distribution information, the historical radiation persistence information, the historical surface temperature information, the historical effective light-receiving area information, the historical incident angle correction coefficient information, the historical open-circuit voltage information, the historical short-circuit current information, the historical fill factor, the historical conversion efficiency information, the historical line resistance information, the historical transmission loss rate information, the historical voltage drop information, the historical charging current information, the historical SOC initial value information, the historical charging and discharging efficiency information, the historical real-time power information, the historical power fluctuation coefficient information, and the historical key load proportion information into a plurality of node features with a transmission order according to the energy transmission order; the node features comprise historical solar radiation node features, historical photovoltaic panel receiving node features, historical photoelectric conversion node features, historical circuit transmission node features, historical energy storage buffer node features, and historical load consumption node features; acquiring solar radiation node sub-models, photovoltaic panel receiving node sub-models, photoelectric conversion node sub-models, circuit transmission node sub-models, energy storage buffer node sub-models, and load consumption node sub-models which are combined in a series mode to form the stable scene photovoltaic prediction sub-model. The historical solar radiation node features, the historical photovoltaic panel receiving node features, the historical photoelectric conversion node features, the historical circuit transmission node features, the historical energy storage buffer node features, and the historical load consumption node features are used to train the solar radiation node submodel, the photovoltaic panel receiving node submodel, the photoelectric conversion node submodel, the circuit transmission node submodel, the energy storage buffer node submodel, and the load consumption node submodel, so as to obtain the trained stable scene photovoltaic prediction submodel.

4. The mobile solar photovoltaic power source energy storage method of claim 3, wherein, The historical solar radiation node features, the historical photovoltaic panel receiving node features, the historical photoelectric conversion node features, the historical circuit transmission node features, the historical energy storage buffer node features, and the historical load consumption node features are used to train the solar radiation node submodel, the photovoltaic panel receiving node submodel, the photoelectric conversion node submodel, the circuit transmission node submodel, the energy storage buffer node submodel, and the load consumption node submodel, so as to obtain the trained stable scene photovoltaic prediction submodel. The historical radiation intensity information, the historical spectral distribution information, the historical radiation persistence information, and the actual value of atmospheric attenuation loss are preprocessed, so as to obtain the preprocessed training solar radiation node submodel data; The solar radiation node submodel is obtained, and the solar radiation node submodel is a multiple linear regression model; The solar radiation node submodel is trained by using the preprocessed training solar radiation node submodel data.

5. The mobile solar photovoltaic power source energy storage method of claim 4, wherein, The historical surface temperature information, the historical effective light receiving area information, and the historical incident angle correction coefficient information are preprocessed, so as to obtain the preprocessed training photovoltaic panel receiving node submodel data; The effective energy data of the solar radiation node output by the solar radiation node submodel obtained in the training of the solar radiation node submodel is obtained; The photovoltaic panel receiving node submodel is obtained, and the photovoltaic panel receiving node submodel is a linear regression model; The photovoltaic panel receiving node submodel is trained by using the preprocessed training photovoltaic panel receiving node submodel data and the effective energy data of the solar radiation node. The historical solar radiation node features, the historical photovoltaic panel receiving node features, the historical photoelectric conversion node features, the historical circuit transmission node features, the historical energy storage buffer node features, and the historical load consumption node features are used to train the solar radiation node submodel, the photovoltaic panel receiving node submodel, the photoelectric conversion node submodel, the circuit transmission node submodel, the energy storage buffer node submodel, and the load consumption node submodel, so as to obtain the trained stable scene photovoltaic prediction submodel.

6. The mobile solar photovoltaic power source energy storage method of claim 5, wherein, ​ Preprocess the historical photoelectric conversion node feature data to obtain preprocessed training photoelectric conversion node sub-model data; Obtain the output energy of the historical photovoltaic panel receiving node feature output by the photovoltaic panel receiving node sub-model obtained in the training of the photovoltaic panel receiving node sub-model; Obtain the photoelectric conversion node sub-model, which is a multiple linear regression model; Train the photoelectric conversion node sub-model by using the preprocessed training photoelectric conversion node sub-model data and the output energy of the historical photovoltaic panel receiving node feature.

7. The mobile solar photovoltaic power source energy storage method of claim 6, wherein, The training of the solar radiation node sub-model, the photovoltaic panel receiving node sub-model, the photoelectric conversion node sub-model, the circuit transmission node sub-model, the energy storage buffer node sub-model, and the load consumption node sub-model using the historical solar radiation node feature, the historical photovoltaic panel receiving node feature, the historical photoelectric conversion node feature, the historical circuit transmission node feature, the historical energy storage buffer node feature, and the historical load consumption node feature further comprises: Preprocess the historical circuit transmission node feature data to obtain preprocessed training circuit transmission node sub-model data; Obtain the output energy of the historical photoelectric conversion node feature output by the photoelectric conversion node sub-model obtained in the training of the photoelectric conversion node sub-model; Obtain the circuit transmission node sub-model, which is a multiple linear regression model; Train the circuit transmission node sub-model by using the preprocessed training circuit transmission node sub-model data and the output energy of the historical photoelectric conversion node feature.

8. The mobile solar photovoltaic power source energy storage method of claim 7, wherein, The training of the solar radiation node sub-model, the photovoltaic panel receiving node sub-model, the photoelectric conversion node sub-model, the circuit transmission node sub-model, the energy storage buffer node sub-model, and the load consumption node sub-model using the historical solar radiation node feature, the historical photovoltaic panel receiving node feature, the historical photoelectric conversion node feature, the historical circuit transmission node feature, the historical energy storage buffer node feature, and the historical load consumption node feature further comprises: Preprocess the historical energy storage buffer node feature data to obtain preprocessed training energy storage buffer node sub-model data; Obtain the output energy of the historical circuit transmission node feature output by the circuit transmission node sub-model obtained in the training of the circuit transmission node sub-model; Obtain the energy storage buffer node sub-model, which is a multiple linear regression model; Train the energy storage buffer node sub-model by using the preprocessed training energy storage buffer node sub-model data and the output energy of the historical photoelectric conversion node feature.

9. The mobile solar photovoltaic power source energy storage method of claim 8, wherein, The use history solar radiation node feature, the history photovoltaic panel receiving node feature, the history photoelectric conversion node feature, the history circuit transmission node feature, the history energy storage buffer node feature, and the history load consumption node feature are respectively used to train a solar radiation node submodel, a photovoltaic panel receiving node submodel, a photoelectric conversion node submodel, a circuit transmission node submodel, an energy storage buffer node submodel, and a load consumption node submodel, so as to obtain a trained stable scene photovoltaic prediction submodel, and the trained stable scene photovoltaic prediction submodel further comprises: The historical load consumption node feature data is preprocessed, so as to obtain preprocessed training load consumption node submodel data; The output energy of the historical energy storage buffer node feature output by the energy storage buffer node submodel obtained in the training of the energy storage buffer node submodel is obtained; A load consumption node submodel is obtained, and the load consumption node submodel is a multiple linear regression model; The load consumption node submodel is trained by using the preprocessed training load consumption node submodel data and the output energy of the historical energy storage buffer node feature.

10. A mobile solar photovoltaic power source energy storage device, characterized in that, The mobile solar photovoltaic power supply energy storage device comprises: a basic information acquisition module, which is used to acquire power information, battery temperature information, load type information, meteorological data, and solar basic information; a feature matrix acquisition module, which is used to construct a multi-dimensional real-time feature matrix according to the power information, the battery temperature information, the load information, the meteorological data, and the solar basic information; a model acquisition module, which is used to acquire a trained scene classification model, a trained stable scene photovoltaic prediction submodel, a trained complex scene photovoltaic prediction submodel, and a trained load prediction model; a scene label acquisition module, which is used to input the multi-dimensional real-time feature matrix into the trained scene classification model, so as to acquire a current scene label; a photovoltaic power prediction module, which is used to select the trained stable scene photovoltaic prediction submodel or the trained complex scene photovoltaic prediction submodel as a current scene model according to the current scene label, and input the multi-dimensional real-time feature matrix into the current scene model, so as to acquire a predicted photovoltaic power; a load prediction module, which is used to input the multi-dimensional real-time feature matrix into the trained load prediction model, so as to acquire an output load prediction value.

Images