A smart lighting control system, method and server based on artificial intelligence

Abstract

The present invention relates to the field of artificial intelligence and big data technology, and discloses an intelligent lighting control system based on artificial intelligence, comprising a lighting recording module, a lighting control module, and a synchronous start-up module; the lighting recording module is used to shoot a video of a lighting area with the lighting light turned on, perform frame extraction processing on the shot video, generate a regional behavior record, and adjust the frame extraction interval of the shot video in real time; the lighting control module is used to control the lighting of the lighting area; the frame extraction interval of the video is adjusted in real time according to the production behavior in the lighting area, the lights in the area that does not need lighting are turned off in time, and a synchronous start-up module is set to perform intelligent analysis on the surrounding area of ​​the lighting area, so that there is no need to turn on the lighting switches at different positions.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and big data technology, and more specifically, to an artificial intelligence-based smart lighting control system, method and server. Background Art

[0002] At present, the lighting system in the factory building needs to provide large-area lighting. The lighting in the factory building needs to be controlled by multiple lighting switches. Usually, the lighting switches are set in different positions. When a certain area in the factory building is in production, the lighting switch in the corresponding area needs to be turned on. When a certain area in the factory building is not in production, the lighting switch in the corresponding position needs to be turned off. However, the lighting switches in the factory building are often forgotten to be turned off or not turned on and off in time. Due to the original circuit design, each time the lighting needs to be turned on, the corresponding personnel may need to go to the lighting switches in different positions to turn on the lights. This control method is not convenient. Summary of the invention

[0003] The present invention provides an intelligent lighting control system based on artificial intelligence to solve the technical problems in related technologies.

[0004] The present invention provides an intelligent lighting control system based on artificial intelligence, including a lighting recording module, a lighting control module, and a synchronous start module;

[0005] The lighting recording module is used to shoot a video of the lighting area when the lighting light is turned on, perform frame extraction processing on the shot video, generate a regional behavior record, and adjust the frame extraction interval of the shot video in real time;

[0006] The lighting control module is used to control the lighting of the lighting area.

[0007] Furthermore, the lighting control module also includes a parameter adjustment module, which generates lighting control parameters of the lighting area based on a parameter generation model, and the parameter generation model includes:

[0008] The parameter generation model inputs graph structure data, which includes the following parts: Node: Each lighting area, equipment, regional environment, and time are regarded as a node;

[0009] Node Characteristics: Each node has a characteristic.

[0010] Get the input node features by:

[0011] Collect data: area / device data, environment data, time data; if the data of a node includes multiple modalities, generate feature vectors through feature engineering respectively and then concatenate them as node features of the node;

[0012] The nodes of two adjacent lighting areas are connected, the nodes of the regional environment are connected to the nodes of the lighting area, the equipment is connected to the nodes of the lighting area where it is located, the time is connected to all other nodes, and edges are built between the connected nodes;

[0013] The calculation formula of the parameter generation model is as follows:

[0014] Messaging:

[0015]

[0016] Where m i←j is the message sent by node j to node i, h i and h j are the current representations of nodes i and j respectively, is a message function, a multilayer perceptron or other neural network.

[0017] Message aggregation:

[0018] m i =ρ ← ({m i←j ∣j∈N(i)})

[0019] Where m i is the aggregation of all messages received by node i, N(i) is the set of neighbor nodes of node i, ρ ← It is a message aggregation function, commonly used ones include average, sum or maximum value.

[0020] Node Update:

[0021]

[0022] where h i ' is node i according to message m i The updated new representation, It is a node update function, usually a multi-layer perceptron or a gated recurrent unit.

[0023] Repeat K iterations to obtain a higher-order node representation. The node representation after iteration is Input to the prediction layer of downstream tasks;

[0024] Prediction layer:

[0025]

[0026] where f pred is the prediction function, N represents the set of all nodes, θ represents vector concatenation, Represents an output vector, which is represented by a digital sequence that can be decoded into specific control parameters.

[0027] Furthermore, the steps for decoding the digital sequence are:

[0028] The output sequence is separated into several blocks by "_", and the digital blocks are parsed according to the pre-defined encoding rules.

[0029] Furthermore, the lighting control module is used to obtain the lighting control value of the lighting area in real time, and determine whether to turn off the lighting light of the lighting area according to the comparison result between the lighting control value and its threshold value;

[0030] When the lighting lights in the lighting area are turned on, the synchronous turning-on module determines whether to turn on the lighting lights in other lighting areas synchronously.

[0031] Furthermore, the lighting environment is divided into multiple lighting areas. When the lighting in the lighting area is turned on, a video of the lighting area is shot to generate a regional video in real time. The regional video is framed based on a preset frame extraction interval. After frame extraction, a regional image is obtained and marked as SEi, where i is the sequential number of the regional image, i=1, 2, ..., n. A regional behavior record is generated synchronously, and the regional behavior record includes the regional behavior value and the regional image time.

[0032] Furthermore, the frame extraction interval of the captured video is adjusted in real time, specifically: a regional behavior threshold is set, and when the regional behavior value of the regional behavior record is less than the regional behavior threshold, the preset frame extraction interval of the regional video is shortened, and the number of shortened frame extraction times is increased by one.

[0033] Furthermore, the regional behavior model of the lighting area is obtained in the following manner: collecting multiple regional images of the lighting area, extracting the production action data of the production personnel in the regional images, assigning different labels to the regional images according to the production action data of the production personnel, the labels are the regional behavior values, dividing the regional images into a training set and a verification set according to the production action data of the production personnel, and obtaining the regional behavior model of the lighting area through iterative training of a neural network. The numerical range of the regional behavior value is (0 to 5). The larger the numerical value of the regional behavior value, the closer the action of the production personnel in the lighting area is to normal production behavior, and vice versa.

[0034] Furthermore, when the lighting in the lighting area is turned on, the lighting area is marked as the lighting center area, and a circle is drawn with a preset radius with the position of the ambient lighting center area as the center to obtain a synchronous control range, and the remaining lighting areas within the synchronous control range are marked as synchronous analysis areas;

[0035] Obtain the total number of stable lighting times and lighting frequency of the lighting center area and mark them as [A1, A2]. Obtain the total number of stable lighting times and lighting frequency of the synchronous analysis area and mark them as [B1, B2]. Use the cosine similarity algorithm to obtain the synchronous lighting values ​​of the lighting center area and the synchronous analysis area.

[0036] Set the synchronous lighting threshold. When the synchronous lighting value of the lighting center area and the synchronous analysis area is greater than or equal to the synchronous lighting threshold, the lighting of the synchronous analysis area is synchronously turned on;

[0037] The total number of lighting stability times is obtained in the following manner: obtain all lighting records of the lighting area, calculate the time difference between the lighting end time and the lighting start time of the lighting record to obtain the lighting duration, sort all lighting durations in the order of the corresponding lighting records, calculate the difference between two adjacent lighting durations after sorting and take the absolute value to obtain the lighting time difference, set the lighting threshold time difference, when the lighting time difference is greater than or equal to the lighting threshold time difference, do not perform corresponding processing, when the lighting time difference is less than the lighting threshold time difference, increase the number of lighting stability times by one, sum the number of lighting stability times, and obtain the total number of lighting stability times of the lighting area;

[0038] The lighting frequency is obtained by the following method: obtain all lighting records in the lighting center area, mark the total number of lighting records as FSJ, sort all lighting records in order, calculate the time difference between the lighting start time of the next lighting record after sorting and the lighting end time of the previous lighting record, and obtain the lighting interval, sum up all lighting intervals and take the average value to obtain the average lighting interval, and mark it as GCM, and use the formula The lighting frequency EHS of the lighting area is obtained, where b1 is the total number coefficient of lighting records and b2 is the average lighting interval coefficient.

[0039] The present invention provides an intelligent lighting control method based on artificial intelligence, comprising the following steps:

[0040] Step 1: Shoot a video of the lighting area with the lighting on, perform frame extraction on the video, generate a regional behavior record, and adjust the frame extraction interval of the video in real time;

[0041] Step 2: Obtain the lighting control value of the lighting area in real time, and determine whether to turn off the lighting in the lighting area according to the comparison result between the lighting control value and the threshold value;

[0042] Step 3: When the lighting in the lighting area is turned on, determine whether to turn on the lighting in other lighting areas synchronously.

[0043] The present invention provides an artificial intelligence-based smart lighting control server for executing the aforementioned artificial intelligence-based smart lighting control method.

[0044] The beneficial effects of the present invention are as follows: the method of the present invention can divide the lighting environment into lighting areas, perform zoned lighting light analysis and control, not only dynamically analyze whether the lighting lights need to be turned off, but also analyze the relevance of turning on the lighting tubes, and optimize the lighting control method of the lighting area;

[0045] Set up a lighting recording module and a lighting control module, and conduct video monitoring of the lighting area. Adjust the video frame interval in real time according to the production behavior in the lighting area, turn off the lights in areas that do not need lighting in time, set up a synchronous start module, and conduct intelligent analysis of the surrounding areas of the lighting area. There is no need to turn on the lighting switches at different locations. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flowchart of the present invention;

[0047] Figure 2 A flowchart for determining whether to turn off the lighting in the lighting area;

[0048] Figure 3 A flowchart for determining whether to turn on the lighting in other lighting areas synchronously. DETAILED DESCRIPTION

[0049] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0050] At least one embodiment of the present invention discloses an intelligent lighting control method based on artificial intelligence, comprising the following steps:

[0051] Step 1: Shoot a video of the lighting area with the lighting on, perform frame extraction on the video, generate a regional behavior record, and adjust the frame extraction interval of the video in real time;

[0052] Step 2: Obtain the lighting control value of the lighting area in real time, and determine whether to turn off the lighting in the lighting area according to the comparison result between the lighting control value and the threshold value;

[0053] Step 3: When the lighting in the lighting area is turned on, determine whether to turn on the lighting in other lighting areas synchronously.

[0054] Through steps one and two, the lighting environment is divided into lighting areas, and video monitoring is performed on the lighting areas. The frame interval of the video is adjusted in real time according to the production behavior in the lighting areas, and the lights in the areas that do not need lighting are turned off in time. Through step three, the surrounding areas of the lighting area are intelligently analyzed to intelligently determine whether the lights in the surrounding areas need to be turned on synchronously. There is no need to turn on the lighting switches at different locations, thereby optimizing the light control method of the lighting area.

[0055] At least one embodiment of the present invention discloses an artificial intelligence-based intelligent lighting control system, including a lighting recording module, a lighting control module and a synchronous start module, wherein:

[0056] Lighting recording module: Divide the lighting environment (such as factory buildings) into multiple lighting areas. When the lighting in the lighting area is turned on (whether it is turned on manually or automatically by the system), a video of the lighting area is shot, and the regional video is generated in real time. Based on the preset frame extraction interval (the frame extraction interval is to extract a frame of the video every specified number of seconds), the regional video is frame extracted, and the regional image is obtained after frame extraction and marked as SEi, where i is the sequential number of the regional image, i=1, 2, ..., n. The regional behavior record is generated synchronously, and the regional behavior record includes the regional behavior value and the regional image time (the time corresponding to the regional image).

[0057] The lighting area can be divided according to different functional areas of the bright environment. For example, when there are tire production areas, frame production areas, and engine production areas in the automobile production plant, the automobile production plant can be divided into three lighting areas. The same area of ​​the lighting environment can be divided into multiple lighting areas, and the division method of the lighting area is not the focus of the present invention.

[0058] The regional behavior value of the regional behavior record is obtained by the following method: inputting the regional image time of the regional behavior record into the regional behavior model of the lighting area, and the regional behavior model of the lighting area outputs the regional behavior value. One lighting area corresponds to one regional behavior model.

[0059] The regional behavior model of the lighting area is obtained by the following method: collecting multiple regional images of the lighting area, extracting the production action data of the production personnel in the regional images, assigning different labels to the regional images according to the production action data of the production personnel, and the labels are the regional behavior values, dividing the regional images into training sets and verification sets according to the production action data of the production personnel, and obtaining the regional behavior model of the lighting area through iterative training of the neural network. The value range of the regional behavior value is (0-5), and the larger the value of the regional behavior value, the closer the production personnel's action in the lighting area is to normal production behavior, and the smaller the value of the regional behavior value is, the opposite is true.

[0060] Set the regional behavior threshold. The regional behavior threshold is the system setting threshold and can be modified according to actual needs.

[0061] When the regional behavior value of the regional behavior record is greater than or equal to the regional behavior threshold, no corresponding processing is performed.

[0062] When the regional behavior value of the regional behavior record is less than the regional behavior threshold, the preset frame extraction interval of the regional video is shortened (the specific shortening range is adjusted according to actual needs), and the number of shortened frame extraction times is increased by one.

[0063] Lighting control module: obtain the lighting control value of the lighting area in real time and set the lighting control threshold. The lighting control threshold is a system preset threshold and can be modified according to actual experimental needs.

[0064] When the lighting control value of a lighting area is greater than or equal to the lighting control threshold, the lighting in that lighting area will be turned off. When the lighting in that lighting area is turned off (whether actively turned off or automatically turned off by the system), a lighting record for that lighting area will be generated synchronously. The lighting record includes the lighting start time and the lighting end time.

[0065] When the lighting control value of the lighting area is less than the lighting control threshold, no corresponding processing is performed.

[0066] The lighting control value of the lighting area is obtained by:

[0067] Obtain all regional behavior records of the lighting area in the lighting-on state, sum up the regional behavior values ​​of all regional behavior records and take the average value to obtain the average regional behavior value, which is marked as PZX, sum up the number of shortened frame extractions of the lighting area in the lighting-on state, obtain the total number of shortened frame extractions, and mark it as TYB, and use the formula The lighting control value LHC of the lighting area is obtained, wherein a1 is the average area value coefficient, a2 is the coefficient for shortening the total number of frame extractions, the value of a1 is 0.99, and the value of a2 is 0.72.

[0068] By setting up a lighting recording module and a lighting control module, video monitoring of the lighting area can be carried out and lights in areas that do not need lighting can be turned off in time.

[0069] Synchronous start-up module: When the lighting in the lighting area is turned on, the lighting area is marked as the lighting center area. With the position of the ambient lighting center area as the center, a circle with a preset radius is drawn (the preset radius can be adjusted according to actual needs) to obtain the synchronous control range. The remaining lighting areas within the synchronous control range are marked as synchronous analysis areas.

[0070] The total number of stable lighting times and lighting frequency of the lighting center area are obtained and marked as [A1, A2]. The total number of stable lighting times and lighting frequency of the synchronous analysis area are obtained and marked as [B1, B2]. The cosine similarity algorithm is used to obtain the synchronous lighting values ​​of the lighting center area and the synchronous analysis area.

[0071] Set the synchronous lighting threshold. The synchronous lighting threshold is the system setting threshold and can be modified according to actual needs.

[0072] When the synchronous lighting values ​​of the lighting center area and the synchronous analysis area are greater than or equal to the synchronous lighting threshold, the lighting lights of the synchronous analysis area are synchronously turned on.

[0073] When the synchronous illumination value of the lighting center area and the synchronous analysis area is less than the synchronous illumination threshold, no corresponding processing is performed.

[0074] For example: The total number of stable lighting times in the lighting center area is 5, the lighting frequency in the lighting center area is 13, the total number of stable lighting times in the synchronous analysis area is 8, and the lighting frequency in the synchronous analysis area is 17.

[0075] The total number of stable lighting times is obtained in the following way: obtain all lighting records of the lighting area, calculate the time difference between the lighting end time and the lighting start time of the lighting record to obtain the lighting duration, sort all lighting durations in sequence according to the order of the corresponding lighting records, calculate the difference between the two adjacent lighting durations after sorting and take the absolute value to obtain the lighting time difference, set the lighting threshold time difference, the lighting threshold time difference is the system setting threshold, which can be modified according to actual needs. When the lighting time difference is greater than or equal to the lighting threshold time difference, no corresponding processing is performed. When the lighting time difference is less than the lighting threshold time difference, the number of stable lighting times is increased by one, and the number of stable lighting times is summed to obtain the total number of stable lighting times for the lighting area.

[0076] The lighting frequency is obtained by the following method: obtain all lighting records in the lighting center area, mark the total number of lighting records as FSJ, sort all lighting records in order, calculate the time difference between the lighting start time of the next lighting record after sorting and the lighting end time of the previous lighting record, and obtain the lighting interval, sum up all lighting intervals and take the average value to obtain the average lighting interval, and mark it as GCM, and use the formula The lighting frequency EHS of the lighting area is obtained, where b1 is the total number coefficient of lighting records, b2 is the average lighting interval coefficient, the value of b1 is 0.59, and the value of b2 is 0.52.

[0077] A synchronous start-up module is set up to perform intelligent analysis of the surrounding area of ​​the lighting area. There is no need to turn on the lighting switches in different locations, thus optimizing the light control method of the lighting area.

[0078] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0079] At least one embodiment of the present invention discloses an artificial intelligence-based intelligent lighting control system, which also includes a parameter adjustment module, which generates lighting control parameters of the lighting area based on a parameter generation model, and the parameter generation model includes:

[0080] The parameter generation model inputs graph structure data, which includes the following parts: Node: Each lighting area, equipment, regional environment, and time are regarded as a node.

[0081] Node characteristics: Each node has a characteristic, such as area type (office, corridor, etc.), area size, equipment type (light, curtain, etc.), etc.

[0082] Get the input node features by:

[0083] Collect data: Area / equipment data: Extract basic information of areas and equipment, such as type, area, location, etc., from data sources such as building information model (BIM) and Internet of Things (IoT) device metadata.

[0084] Environmental data: Collect data about people's activities.

[0085] Time data: Extract time features such as hours, working days / rest days, etc.

[0086] The node features of the corresponding nodes are generated by the collected data, for example, the node features of the lighting area are generated by the area data.

[0087] If the data of a node includes multiple modes, the feature vectors are generated through feature engineering respectively and then concatenated as the node features of the node;

[0088] The nodes of two adjacent lighting areas are connected, the nodes of the regional environment are connected to the nodes of the lighting area, the equipment is connected to the nodes of the lighting area where it is located, the time is connected to all other nodes, and edges are built between the connected nodes.

[0089] The calculation formula of the parameter generation model is as follows:

[0090] Messaging:

[0091]

[0092] Where m i←j is the message sent by node j to node i, h i and h j are the current representations of nodes i and j respectively, is a message function, a multilayer perceptron or other neural network.

[0093] Message aggregation:

[0094] m i =ρ ← ({m i←j ∣j∈N(i)})

[0095] Where m i is the aggregation of all messages received by node i, N(i) is the set of neighbor nodes of node i, ρ ← It is a message aggregation function, commonly used ones include average, sum or maximum value.

[0096] Node Update:

[0097]

[0098] where h i 'Node i according to message m i The updated new representation, It is a node update function, usually a multi-layer perceptron or a gated recurrent unit.

[0099] The above process can be repeated K times on the entire graph to obtain a higher-order node representation. Finally, the node representation Input to a prediction layer for downstream tasks, such as a multi-layer perceptron or a linear layer, to predict lighting control parameters:

[0100]

[0101] where f predis the prediction function, N represents the set of all nodes, θ represents vector concatenation, Represents an output vector, which is represented by a digital sequence that can be decoded into specific control parameters.

[0102] For the control parameters of the lighting area, a typical example may include the following main parts:

[0103] Schedule

[0104] Opening time: 07:00

[0105] Closing time: 19:00

[0106] Brightness Level

[0107] Daytime: 80%

[0108] Night: 50%

[0109] Partition Settings

[0110] Area 1: Open

[0111] Area 2: Closed

[0112] Area 3: Open

[0113] Other settings

[0114] Sensor sensitivity: Medium

[0115] Delay time: 5 minutes

[0116] These parameters can be combined into a sequence, such as:

[0117] "07:0019:000.80.51012300"

[0118] The meaning of the digital code can be predefined.

[0119] For the output of the model, a sequence generation method can be used. For example, for the above parameter sequence, the model can generate:

[0120] "0700_1900_80_50_101_2_300"

[0121] This is a simple serialized representation. When deployed, a decoding module can be written to parse this sequence into specific control parameter settings.

[0122] The decoding steps can be:

[0123] Separate the output sequence into several blocks by "_"

[0124] The first two blocks decode to time, for example "0700" corresponds to 07:00

[0125] The following digital blocks are parsed according to predefined encoding rules

[0126] Send the parsed parameter settings to the lighting control system for execution

[0127] This is just one possible way to represent and decode parameters. In practice, some standardized forms can be followed to make encoding and decoding easier to implement. At the same time, mechanisms such as check bits can be introduced into the parameter sequence to enhance fault tolerance.

[0128] In general, through reasonable serialization representation and encoding and decoding methods, it is possible to achieve seamless connection from model output to control parameter setting.

[0129] The goal of designing a loss function is to measure the gap between the predicted lighting control parameters and the actual optimal parameters. It is necessary to consider the characteristics of different parameter types (such as time, brightness, switch status, etc.) and choose an appropriate measurement method.

[0130] Time parameter: For the time parameter, the mean square error (MSE) can be used as the loss function. MSE calculates the average of the sum of the squares of the differences between two time points. For example, assuming that the model predicts the start time to be t1 = 07:30, and the actual optimal time is t2 = 07:30, the loss function can be defined as:

[0131] loss=(t1-t2)*2

[0132] Brightness parameter: For brightness parameter, MSE can also be used to calculate the sum of squares of the difference between the predicted brightness value and the actual optimal brightness value. For example:

[0133] loss=(predicted b rightness-optimal b rightness)*2

[0134] Switch state parameters: For the switch state (such as the switch of area 1, area 2, etc.), a 0-1 loss can be used. If the predicted switch state is consistent with the actual optimal state, the loss is 0; if not, the loss is 1. For example:

[0135] loss = int(predicted s tate! =optimal s tate)

[0136] Comprehensive loss function: By combining the above loss functions, we can design a composite loss function, such as setting different weights according to the importance of parameters by weighted average. For example:

[0137] loss=weight t ime*loss t ime+weight b rightness*loss b rightness

[0138] +weight s witch*loss s witch

[0139] Where weight t ime、weight b rightness、weight s witch are weights set according to the importance of the parameters.

[0140] Consider constraints: If the control parameters have specific constraints (such as brightness cannot exceed 100%, time cannot exceed the day boundary), introduce a penalty term in the loss function to ensure that the parameters meet these constraints.

[0141] In one embodiment of the present invention, for the loss function of the time parameter, the mean square error (MSE) is a good choice. However, since time has a periodic characteristic, it is also possible to consider using a cyclic loss function, such as:

[0142]

[0143] where t pred and t opt are the predicted time and expected time, in hours. TT is a constant representing the number of hours in a day (usually 24). This loss function can handle the span problem of night and morning well.

[0144] For the brightness parameter, MSE is a standard choice. However, sometimes you may also consider using a smooth L1 loss function:

[0145]

[0146] Where x is the predicted brightness and y is the expected brightness. This loss function has less impact on outliers and can enhance the robustness of the model.

[0147] As for the switch state, the 0-1 loss function you mentioned is reasonable, but sometimes you can also consider introducing some softening, such as using sigmoid cross entropy loss:

[0148]

[0149] Where y is the expected value of the switch state (0 or 1), σ is the sigmoid function, and x is the model output (real number).

[0150] The comprehensive loss function can be a weighted sum of these loss terms:

[0151]

[0152] where λ i It is the corresponding weight coefficient, which can be adjusted according to the preference for different types of loss weights. The value ranges from 0 to 1, and the sum is 1.

[0153] Evaluation metrics: When training a model, in addition to the loss function, you should also consider using evaluation metrics (such as RMSE, MAE, etc.) to evaluate the model's prediction results to ensure that the model's performance on both the training and validation sets reaches the expected performance level.

[0155] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. An intelligent lighting control system based on artificial intelligence, characterized in that: It includes lighting recording module, lighting control module and synchronous opening module; The lighting recording module is used to shoot a video of the lighting area when the lighting light is turned on, perform frame extraction processing on the shot video, generate a regional behavior record, and adjust the frame extraction interval of the shot video in real time; The lighting control module is used to control the lighting of the lighting area.

2. According to claim 1, the intelligent lighting control system based on artificial intelligence is characterized in that: The lighting control module also includes a parameter adjustment module, which generates lighting control parameters of the lighting area based on a parameter generation model, and the parameter generation model includes: The parameter generation model inputs graph structure data, which includes the following parts: Node: Each lighting area, equipment, regional environment, and time are regarded as a node; Node Features: Each node has a feature; Get the input node features by: Collect data: area / device data, environment data, time data; if the data of a node includes multiple modalities, generate feature vectors through feature engineering respectively and then concatenate them as node features of the node; The nodes of two adjacent lighting areas are connected, the nodes of the regional environment are connected to the nodes of the lighting area, the equipment is connected to the nodes of the lighting area where it is located, the time is connected to all other nodes, and edges are built between the connected nodes; The calculation formula of the parameter generation model is as follows: Messaging: Where m i←j is the message sent by node j to node i, h i and h j are the current representations of nodes i and j respectively, is the message function; Message aggregation: m i =ρ ← ({m i←j ∣j∈N(i)}) Where m i is the aggregation of all messages received by node i, N(i) is the set of neighbor nodes of node i, ρ ← is the message aggregation function; Node Update: where h i 'Node i according to message m i The updated new representation, is the node update function; Repeat K iterations to obtain a higher-order node representation. The node representation after iteration is Input to the prediction layer of downstream tasks; Prediction layer: where f pred is the prediction function, N represents the set of all nodes, θ represents vector concatenation, Represents an output vector, which is represented by a digital sequence that can be decoded into specific control parameters.

3. The intelligent lighting control system based on artificial intelligence according to claim 2 is characterized in that: The steps to decode the number sequence are: The output sequence is separated into several blocks by "_", and the digital blocks are parsed according to the pre-defined encoding rules.

4. The intelligent lighting control system based on artificial intelligence according to claim 1 is characterized in that: The lighting control module is used to obtain the lighting control value of the lighting area in real time, and determine whether to turn off the lighting in the lighting area according to the comparison result between the lighting control value and its threshold value; When the lighting lights in the lighting area are turned on, the synchronous turning-on module determines whether to turn on the lighting lights in other lighting areas synchronously.

5. The intelligent lighting control system based on artificial intelligence according to claim 1 is characterized in that: The lighting environment is divided into multiple lighting areas. When the lighting in the lighting area is turned on, a video of the lighting area is shot to generate a regional video in real time. The regional video is framed based on a preset frame extraction interval. After frame extraction, a regional image is obtained and marked as SEi, where i is the sequential number of the regional image, i=1, 2, ..., n. A regional behavior record is generated synchronously, and the regional behavior record includes the regional behavior value and the regional image time.

6. The intelligent lighting control system based on artificial intelligence according to claim 5 is characterized in that: The frame extraction interval of the captured video is adjusted in real time, specifically: a regional behavior threshold is set. When the regional behavior value of the regional behavior record is less than the regional behavior threshold, the preset frame extraction interval of the regional video is shortened, and the number of shortened frame extraction times is increased by one.

7. The intelligent lighting control system based on artificial intelligence according to claim 5 is characterized in that: The regional behavior model of the lighting area is obtained in the following way: collecting multiple regional images of the lighting area, extracting the production action data of the production personnel in the regional images, assigning different labels to the regional images according to the production action data of the production personnel, and the labels are the regional behavior values. The regional images are divided into a training set and a verification set according to the production action data of the production personnel, and the regional behavior model of the lighting area is obtained through iterative training of the neural network. The numerical range of the regional behavior value is (0~5). The larger the numerical value of the regional behavior value, the closer the action of the production personnel in the lighting area is to the normal production behavior, and the smaller the numerical value of the regional behavior value is, the opposite is true.

8. The intelligent lighting control system based on artificial intelligence according to claim 1 is characterized in that: When the lighting in the lighting area is turned on, the lighting area is marked as the lighting center area, and a circle is drawn with a preset radius with the position of the ambient lighting center area as the center to obtain the synchronous control range, and the remaining lighting areas within the synchronous control range are marked as synchronous analysis areas; Obtain the total number of stable lighting times and lighting frequency of the lighting center area and mark them as [A1, A2]. Obtain the total number of stable lighting times and lighting frequency of the synchronous analysis area and mark them as [B1, B2]. Use the cosine similarity algorithm to obtain the synchronous lighting values ​​of the lighting center area and the synchronous analysis area. Set the synchronous lighting threshold. When the synchronous lighting value of the lighting center area and the synchronous analysis area is greater than or equal to the synchronous lighting threshold, the lighting of the synchronous analysis area is synchronously turned on; The total number of lighting stability times is obtained in the following manner: obtain all lighting records of the lighting area, calculate the time difference between the lighting end time and the lighting start time of the lighting record to obtain the lighting duration, sort all lighting durations in the order of the corresponding lighting records, calculate the difference between two adjacent lighting durations after sorting and take the absolute value to obtain the lighting time difference, set the lighting threshold time difference, when the lighting time difference is greater than or equal to the lighting threshold time difference, do not perform corresponding processing, when the lighting time difference is less than the lighting threshold time difference, increase the number of lighting stability times by one, sum the number of lighting stability times, and obtain the total number of lighting stability times of the lighting area; The lighting frequency is obtained by the following method: obtain all lighting records in the lighting center area, mark the total number of lighting records as FSJ, sort all lighting records in order, calculate the time difference between the lighting start time of the next lighting record after sorting and the lighting end time of the previous lighting record, and obtain the lighting interval, sum up all lighting intervals and take the average value to obtain the average lighting interval, and mark it as GCM, and use the formula The lighting frequency EHS of the lighting area is obtained, where b1 is the total number coefficient of lighting records and b2 is the average lighting interval coefficient.

9. An intelligent lighting control method based on artificial intelligence, applied to an intelligent lighting control system based on artificial intelligence according to any one of claims 1 to 8, characterized in that: The steps include: Step 1: Shoot a video of the lighting area with the lighting on, perform frame extraction on the video, generate a regional behavior record, and adjust the frame extraction interval of the video in real time; Step 2: Obtain the lighting control value of the lighting area in real time, and determine whether to turn off the lighting in the lighting area according to the comparison result between the lighting control value and the threshold value; Step 3: When the lighting in the lighting area is turned on, determine whether to turn on the lighting in other lighting areas synchronously.

10. An intelligent lighting control server based on artificial intelligence, characterized in that: Applicable to executing the method described in claim 9.

Images