Non-intrusive energy storage power station sensing terminal safety monitoring method and system

Abstract

The invention relates to a non-intrusive energy storage power station sensing terminal safety monitoring method and system. The method comprises the steps that non-intrusive power consumption measuring equipment is used for collecting power consumption flow data; extracting an observation sample by using a sliding window technology; building a mixed probability distribution estimation model for describing the working state of the sensing terminal of the energy storage power station; constructing a target function of an end-to-end training mixed probability distribution estimation model; training the hybrid probability distribution estimation model on line by using the observation sample and the target function to obtain a parameter estimation matrix of hybrid probability distribution; and calculating an abnormal score according to the mixed probability distribution estimation model and the parameter estimation matrix to obtain the safety state judgment of the sensing terminal of the energy storage power station. The method breaks through the limitations of a traditional monitoring method in an intrusive mode, a single working condition, dependence on label abnormal samples and the like, the monitoring data is obtained in a non-intrusive mode, only normal samples are used for online training, online monitoring under different working conditions can be adapted, and a practical solution is provided for safety monitoring of the sensing terminal of the energy storage power station.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage power station safety monitoring, and particularly relates to a non-invasive safety monitoring method and system for a sensing terminal of an energy storage power station. Background Art

[0002] With the wide application of energy storage power stations in the energy field, their safe operation is of crucial importance. As a key device for monitoring the operating state of an energy storage system, the safety monitoring of the working state of the sensing terminal of an energy storage power station is directly related to the overall stability and reliability of the energy storage power station. Traditional monitoring methods often require hardware-invasive transformation of the sensing terminal. In actual operation, this not only requires professional technicians to spend a lot of time and effort, but also may damage the original structure and performance of the sensing terminal, thereby affecting the normal operation of the energy storage power station. In addition, traditional monitoring methods are difficult to adapt to the complex and changeable working states of the sensing terminal. When the working state of the sensing terminal changes, such as switching from a low-power consumption mode to a high-power consumption mode, false judgments or missed judgments are likely to occur. In particular, many traditional monitoring methods rely on a labeled abnormal sample data set for training and analysis, but in actual applications, collecting and labeling abnormal sample data is a difficult task. The occurrence of abnormal situations usually has randomness and uncertainty. Obtaining a sufficient number of accurately labeled abnormal samples requires a lot of time, manpower, and material resources. Once a new type of abnormal situation appears, since it is not covered in the training data, the monitoring model may not be able to accurately identify it. Therefore, it is of great practical significance to develop a non-invasive, multi-working-condition-adaptive, and label-abnormal-sample-data-set-independent safety monitoring method for the sensing terminal of an energy storage power station. Summary of the Invention

[0003] Embodiments of the present invention provide a non-invasive safety monitoring method and system for a sensing terminal of an energy storage power station to solve the problems of invasiveness, difficulty in adapting to multiple working conditions, and dependence on labeled abnormal samples existing in the existing monitoring methods, and to realize online monitoring of the safety state of the sensing terminal of the energy storage power station.

[0004] The present invention adopts the following technical solutions: In a first aspect, an embodiment of the present application provides a non-invasive safety monitoring method for a sensing terminal of an energy storage power station, including: Obtaining power consumption flow data of the sensing terminal of the energy storage power station by a non-invasive power consumption measurement device; Extracting observation samples from the flow data by using a sliding window, where the observation samples include working state label encodings and instantaneous power values corresponding to a continuous plurality of timestamps; Building a hybrid probability distribution estimation model for describing the working state of the sensing terminal of the energy storage power station; Construct an objective function for an end-to-end training hybrid probability distribution estimation model, where the objective function includes a working state recognition loss term, a negative log-likelihood term, and a regularization term; Perform end-to-end online training on the hybrid probability distribution estimation model using the observation samples and the objective function to obtain a parameter estimation matrix of the hybrid probability distribution; Extract new observation samples from the flow data, then calculate an anomaly score according to the trained hybrid probability distribution estimation model and the parameter estimation matrix, and further obtain the safety status determination of the current energy storage power station sensing terminal.

[0005] In a second aspect, an embodiment of the present application provides a non-intrusive safety monitoring system for an energy storage power station sensing terminal, including: A flow data acquisition module for acquiring power consumption flow data of an energy storage power station sensing terminal using a non-intrusive power consumption measurement device; A flow data preprocessing module for extracting observation samples from the flow data using a sliding window, where the observation samples include working state label encodings and instantaneous power values corresponding to a continuous plurality of timestamps; A model construction module for building a hybrid probability distribution estimation model describing the working state of an energy storage power station sensing terminal; An objective function construction module for constructing an objective function for end-to-end training of a hybrid probability distribution estimation model, where the objective function includes a working state recognition loss term, a negative log-likelihood term, and a regularization term; A model training module for performing end-to-end online training on the hybrid probability distribution estimation model using the observation samples and the objective function to obtain a parameter estimation matrix of the hybrid probability distribution; An online monitoring module for extracting new observation samples from the flow data, then calculating an anomaly score according to the trained hybrid probability distribution estimation model and the parameter estimation matrix, and further obtaining the safety status determination of the current energy storage power station sensing terminal.

[0006] In a third aspect, an embodiment of the present application provides an electronic device, including: One or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in the first aspect.

[0007] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which computer instructions are stored, and when the instructions are executed by a processor, the steps of the method as described in the first aspect are implemented.

[0008] The technical solutions provided by the embodiments of the present application may include the following beneficial effects: 1. Good compatibility and practicability. The monitoring method of the present invention is non-invasive monitoring, which does not require hardware modification of the sensing terminal of the energy storage power station. Only power consumption data is measured, which can be compatible with different types of sensing terminals, thus greatly reducing the on-site equipment cost and reducing the interference to the in-service system, and improving the compatibility and practicability of the sensing terminal safety monitoring system; 2. Good adaptability and robustness. By modeling the probability distribution of different working states, the present invention can adapt to multiple working conditions at the same time. Even in the face of a combination of working states or abnormal situations that have never appeared before, the potential risks can be evaluated based on the dynamic probability distribution; 3. Support for online iterative upgrade. The present invention adopts an online training method, which can be adjusted according to actual needs such as production lines and production scenarios, and online update the parameters of the mixed probability distribution estimation model and the parameter estimation matrix of the mixed probability distribution to realize the online upgrade of the monitoring system to quickly support the monitoring in new scenarios; 4. Low model development cost and short cycle. The present invention adopts an end-to-end online training method. The model training process only uses normal data, getting rid of the dependence on the labeled abnormal sample data set, reducing the difficulty and cost of data collection, shortening the model development cycle, and realizing rapid development and application; It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit this application. Description of the Drawings

[0009] The drawings here are incorporated into the specification and constitute a part of this specification, showing the embodiments consistent with this application, and are used together with the specification to explain the principles of this application.

[0010] Figure 1 It is a flowchart of the non-invasive energy storage power station sensing terminal safety monitoring method according to the embodiment of the present invention.

[0011] Figure 2 It is a schematic diagram of the power consumption flow data observation sample provided by the embodiment of the present invention.

[0012] Figure 3 It is a schematic diagram of the structure of the working state mixed probability distribution estimation model provided by the embodiment of the present invention.

[0013] Figure 4 It is a safety monitoring result diagram of the energy storage power station sensing terminal provided by the embodiment of the present invention.

[0014] Figure 5 It is a block diagram of the non-invasive energy storage power station sensing terminal safety monitoring system provided by the embodiment of the present invention. Detailed Embodiments

[0015] Exemplary embodiments will be described in detail herein, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0016] The terms used in the present application are for the purpose of describing specific embodiments only and are not intended to limit the present application. The singular forms "a", "the", and "said" used in the present application and the appended claims are also intended to include the plural forms unless the context clearly dictates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0017] Figure 1 A flowchart of a non-intrusive energy storage power station sensing terminal security monitoring method is shown. The method may include the following steps: S1: Obtain the power consumption flow data of the energy storage power station sensing terminal using a non-intrusive power consumption measurement device; S2: Extract observation samples from the flow data using a sliding window. The observation samples include the working state label encodings and instantaneous power values corresponding to a continuous plurality of timestamps; S3: Build a hybrid probability distribution estimation model for describing the working state of the energy storage power station sensing terminal; S4: Construct an objective function for end-to-end training of the hybrid probability distribution estimation model. The objective function includes a working state recognition loss term, a negative log-likelihood term, and a regularization term; S5: Use the observation samples and the objective function to perform end-to-end online training on the hybrid probability distribution estimation model to obtain a parameter estimation matrix of the hybrid probability distribution; S6: Extract new observation samples from the flow data, and then calculate an anomaly score according to the trained hybrid probability distribution estimation model and the parameter estimation matrix, thereby obtaining a determination of the security state of the current energy storage power station sensing terminal.

[0018] As can be seen from the above embodiments, the present invention application obtains the power consumption flow data of the energy storage power station sensing terminal by using a non-invasive power consumption measurement device, and then extracts the observation samples including the working state label code and the instantaneous power value from the flow data by using a sliding window; then constructs a hybrid probability distribution estimation model for describing the working state of the energy storage power station sensing terminal, and constructs an end-to-end training objective function including a working state recognition loss term, a negative log-likelihood term, and a regularization term; based on the observation samples and the objective function, performs end-to-end online training on the hybrid probability distribution estimation model to obtain a parameter estimation matrix of the hybrid probability distribution; finally, extracts new observation samples from the flow data, calculates the anomaly score according to the hybrid probability distribution estimation model and the parameter estimation matrix, and determines the current safety state according to the comparison result between the anomaly score and the preset risk threshold, so as to realize the online safety monitoring of the energy storage power station sensing terminal.

[0019] In terms of technical advantages, the present invention uses a non-invasive measurement device to obtain power consumption flow data, and there is no need to modify the hardware of the sensing terminal throughout the process, greatly reducing the monitoring cost and interference to the system. Different from the existing monitoring methods, the present invention models the hybrid probability distribution of different working states of the energy storage power station sensing terminal, can adapt to various working conditions at the same time, and can evaluate potential risks according to the dynamic probability distribution even in the face of working state combinations or abnormal situations that have never appeared before. It should be noted that the present invention supports online training, and can update the parameters of the hybrid probability distribution estimation model and the parameter estimation matrix of the hybrid probability distribution in real time according to actual needs, continuously optimizing the monitoring effect. In addition, the present invention gets rid of the dependence on the abnormal sample data set, greatly reduces the difficulty of obtaining training data, enables the monitoring system to be built and deployed more quickly and efficiently, and provides a more reliable and timely guarantee for the safe and stable operation of the energy storage power station.

[0020] In the specific implementation of S1: obtain the power consumption flow data of the energy storage power station sensing terminal by using a non-invasive power consumption measurement device; This step may include the following sub-steps: S11: Using the principle of electromagnetic induction, the non-invasive power consumption measurement device senses the change of the magnetic field around the sensing terminal cable through a Rogowski coil, and converts it into an electrical signal, so as to obtain the instantaneous power of the sensing terminal.

[0021] Specifically, the induced electromotive force detected by the sensor is , according to Faraday's law of electromagnetic induction: ; In the formula, is the number of turns of the coil, is the change of the magnetic flux with time ; For the Rogowski coil, its output voltage is related to the measured current The relationship is as follows: ; In the formula, is the mutual inductance coefficient. By performing operations such as integrating the induced electromotive force or output voltage, the corresponding current value is obtained, and then combined with the measured voltage value , the instantaneous power is calculated; S12: Standardize and store the power consumption flow data; Specifically, the non-intrusive power consumption measurement device collects data in real time according to a set frequency, and the obtained flow data is stored in JSON format. The sensing terminal device identifier is , the timestamp is , the working state is , and the instantaneous power is ; Taking the temperature sensors in the energy storage power station used to monitor overheating of batteries, inverters, and energy storage devices in the energy storage system as an example, when the temperature sensors are working normally, they can be divided into an energy-saving low-power sleep state, a medium-power standby state for rapid response measurement, a high-power measurement state for real-time measurement, and an ultra-high-power calibration state for ensuring accuracy according to the power consumption level.

[0022] S13: Transmit the flow data. Transmit the collected data to the subsequent processing module through wireless communication protocols such as low-power Bluetooth, ZigBee, or LoRa. In addition, serial communication can also be used; Serial communication is suitable for scenarios with high requirements for real-time data transmission, short transmission distances, and convenient wiring, such as data interaction between devices within a short distance in the energy storage power station; Standard serial protocols such as RS-232 and RS-485 are adopted to ensure the stability of data transmission, which can complement wireless communication methods to meet different application requirements.

[0023] In the specific implementation of S2: Use a sliding window to extract observation samples from the flow data. The observation samples include the working state label codes and instantaneous power values corresponding to multiple consecutive timestamps; Specifically, when using a sliding window to extract observation samples from the flow data, the length of the sliding window is set to , the step size is set to , , are pre-set positive integers, and the observation sample is: ; In the formula, X represents the set of instantaneous power values, Represents the corresponding set of working state codes. The working states are coded starting from 0. For example, the low-power sleep state code is 0, the medium-power standby state code is 1, the high-power measurement state code is 2, and the ultra-high-power calibration state code is 3; The instantaneous power values in set X and the working state codes at the corresponding index positions in the set correspond one by one, jointly constituting a complete observation sample information; An example of an observation sample is Figure 2 as shown. The left axis is the instantaneous power value, and the right axis is the working state.

[0024] In the specific implementation of S3: Build a hybrid probability distribution estimation model to describe the working state of the energy storage power station sensing terminal; This step may include the following sub-steps: S31: Perform simple preprocessing on the observation samples extracted from the power consumption flow data of the energy storage power station sensing terminal, and then directly use them as model inputs; Specifically, first detect missing values in the observation samples; then for the missing values, fill them with preset fixed values to ensure the data integrity of the observation samples; in order to be able to clearly distinguish from the real data, the fixed value is usually set to -1 because the real instantaneous power value should always be greater than or equal to zero, and using -1 as the filling value helps to identify the missing observation data in the subsequent model training.

[0025] S32: Build a hybrid probability distribution estimation model for the working state of the energy storage power station sensing terminal based on a neural network, which is described by the formula: ; In the formula, is the latent feature vector of the observation sample; is the mixed probability corresponding to different working states, satisfying , is the number of types of working states, represents the probability of belonging to the th working state; is the parameter of the hybrid probability distribution estimation model.

[0026] Specifically, an example of the structure of a simple hybrid probability distribution estimation model is Figure 3As shown, it includes an input layer, a normalization layer, a latent feature vector extraction module, a probability distribution estimation module, and an output layer. Among them, the normalization layer uses LayerNorm to normalize the input observation samples, making the mean of the observation samples 0 and the variance 1, reducing the internal covariate shift, so as to reduce the impact of differences between different samples on the model, enabling the model to have a more stable performance for various input data and improving the generalization ability of the model; the probability normalization layer is implemented using the SoftMax function.

[0027] In the specific implementation of S4: construct the objective function of the end-to-end training hybrid probability distribution estimation model, and the objective function includes a working state recognition loss term, a negative log-likelihood term, and a regularization term; This step may include the following sub-steps: S41: The overall structure of the objective function for end-to-end training is as follows: ; In the formula, is the number of observation samples; is the working state recognition loss term; is the working state soft label of the th observation sample; is the mixed probability of the th observation sample corresponding to different working states; is the negative log-likelihood function; is the latent feature vector of the th observation sample; is the parameter matrix of the mixed probability distribution; is the covariance regularization function; and are the meta-parameters of the mixed probability model. According to engineering experience, and usually perform well; S42: Construct the working state recognition loss term in the objective function. Specifically, use the cross-entropy loss function, and the calculation formula is: ; In the formula, is the ratio of the th observation sample corresponding to the th class of working state, which is calculated by the proportion of the number of the th class of working state in the observation sample, with a maximum value of 1 and a minimum value of 0; is the probability value of the th observation sample corresponding to the th class of working state, with a maximum value of 1 and a minimum value of 0.

[0028] S43: Construct the negative log-likelihood term in the objective function. Considering the universality of the Gaussian mixture distribution, the mixture probability distribution is preset as the Gaussian mixture distribution. The calculation formula for the negative log-likelihood term is as follows: ; In the formula, , , are respectively the mixture probability matrix, the expectation matrix, and the covariance matrix of the th working state. The calculation formulas are as follows: ; S44: Construct the regularization term in the objective function to ensure that the covariance matrix is non-singular. The calculation formula is as follows: ; In the formula, is the dimension of the latent feature vector.

[0029] In the specific implementation of S5: Use the observed samples and the objective function to perform end-to-end online training on the mixture probability distribution estimation model to obtain the parameter estimation matrix of the mixture probability distribution; This step may include the following sub-steps: S51: Update the parameters of the mixture probability distribution estimation model using the gradient descent method. The calculation formula is as follows: ; In the formula, is the learning rate, is the gradient of the objective function at the th iteration, is the parameter of the estimation model at the th iteration; S52: Update the parameter estimation matrix of the mixture probability distribution. Specifically, calculate the mixture probability matrix , the expectation matrix , and the covariance matrix according to the mixture probabilities of the observed samples output by the mixture probability distribution estimation model.

[0030] In the specific implementation of S6: Extract new observed samples from the flow data, then calculate the anomaly score according to the trained mixture probability distribution estimation model and the parameter estimation matrix, and further obtain the safety status determination of the current energy storage power station sensing terminal; This step may include the following sub-steps: S61: Use a sliding window to extract new observed samples from the power consumption flow data . The length of the sliding window is the same as that during online training. The step size can be smaller than the length of the sliding window to achieve overlapping sampling and reduce the risk of missed detection. For example,Figure 4 As shown, the sliding window length is 200 and the step size is 50.

[0031] S62: Calculate the negative log-likelihood of the observation sample according to the working state mixture probability distribution estimation model obtained by online training and the parameter estimation matrix of the mixture probability distribution to obtain the anomaly score of the current observation sample. The anomaly score is calculated by the following formula: : ; In the formula, is the anomaly score of the current observation sample, are the parameters of the trained mixture probability distribution estimation model, represents only returning the latent feature vector of the current observation sample, are respectively the mixture probability matrix, expectation matrix, and covariance matrix estimated by online training.

[0032] S63: Preset a risk threshold . When the calculated anomaly score is greater than the risk threshold , it is determined that the sensing terminal of the energy storage power station is in an unsafe state and an alarm is triggered. As Figure 4 shown, the risk threshold is 0.001.

[0033] To verify the accuracy and effectiveness of the non-intrusive sensing terminal safety monitoring method for the energy storage power station of the present invention, an anomaly injection experiment was performed in this embodiment. The sliding window length was set to 200, the step size was set to 50, and the threshold was set to 0.001. The experimental results are as Figure 4 shown, demonstrating that the method provided by the present invention can identify anomalies in different working states, strongly proving its excellent performance in practical applications. It should be noted that traditional monitoring methods are difficult to accurately identify abnormal situations when faced with frequent switching and changes in working states. By modeling the probability distribution in different working states, the monitoring system of the present invention is not limited to fixed abnormal pattern judgment, but dynamically evaluates according to the probability distribution in different working states. Even in the face of working state combinations or abnormal situations that have never appeared before, it can effectively identify potential risks, greatly enhancing the adaptability to different working states and ensuring reliable safety monitoring of the sensing terminal of the energy storage power station under various working conditions.

[0034] As Figure 5 shown, the present invention also provides a non-intrusive sensing terminal safety monitoring system for an energy storage power station. The system includes: A streaming data acquisition module 1 for acquiring the power consumption streaming data of the sensing terminal of the energy storage power station by using a non-intrusive power consumption measurement device; The streaming data preprocessing module 2 is used to extract observation samples from the streaming data by using a sliding window, and the observation samples include the working state label encodings and instantaneous power values corresponding to a continuous plurality of timestamps; The model construction module 3 is used to construct a hybrid probability distribution estimation model for describing the working state of the energy storage power station sensing terminal; The objective function construction module 4 is used to construct an objective function for end-to-end training of the hybrid probability distribution estimation model, and the objective function includes a working state recognition loss term, a negative log-likelihood term, and a regularization term The model training module 5 is used to perform end-to-end online training on the hybrid probability distribution estimation model by using the observation samples and the objective function to obtain a parameter estimation matrix of the hybrid probability distribution; The online monitoring module 6 is used to extract new observation samples from the streaming data, then calculate an anomaly score according to the trained hybrid probability distribution estimation model and the parameter estimation matrix, and further obtain a safety state determination of the current energy storage power station sensing terminal.

[0035] Regarding the system in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated herein.

[0036] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can refer to the partial descriptions of the method embodiments. The system embodiments described above are only illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of the present application. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0037] Correspondingly, the present application further provides an electronic device, including: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement a non-intrusive safety monitoring method for an energy storage power station sensing terminal as described above.

[0038] Correspondingly, the present application further provides a computer-readable storage medium, on which computer instructions are stored, and when the instructions are executed by a processor, a non-intrusive safety monitoring method for an energy storage power station sensing terminal as described above is implemented.

[0039] Other embodiments of the present application will be readily apparent to those skilled in the art upon consideration of the specification and practice of the disclosure herein. The present application is intended to cover any variations, uses, or adaptations of the present application, which follow the general principles of the present application and include known common knowledge or conventional technical means in the technical field not disclosed in the present application. The specification and examples are only to be considered as exemplary, and the true scope and spirit of the present application are pointed out by the claims.

[0040] It should be understood that the present application is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present application is only limited by the appended claims.

Claims

1. A non-intrusive energy storage power station sensor terminal safety monitoring method, characterized in that: include: Use non-intrusive power consumption measurement equipment to obtain power consumption flow data of energy storage power station sensor terminals; Extracting observation samples from the stream data using a sliding window, wherein the observation samples include working state label codes and instantaneous power values ​​corresponding to a plurality of consecutive timestamps; Build a mixed probability distribution estimation model to describe the working status of the sensor terminal of the energy storage power station; Constructing an objective function of an end-to-end training mixed probability distribution estimation model, wherein the objective function includes a working state recognition loss term, a negative log-likelihood term, and a regularization term; Performing end-to-end online training on the mixed probability distribution estimation model using the observed samples and the objective function to obtain a parameter estimation matrix of the mixed probability distribution; New observation samples are extracted from the stream data, and anomaly scores are calculated based on the trained mixed probability distribution estimation model and the parameter estimation matrix, thereby obtaining a safety status judgment of the current energy storage power station sensor terminal.

2. The method according to claim 1, characterized in that The non-intrusive power consumption measurement device uses a Rogowski coil to sense the magnetic field changes around the sensor terminal cable and converts it into an electrical signal, thereby obtaining the sensor terminal power consumption flow data, the flow data including the working state and instantaneous power.

3. The method according to claim 1, characterized in that The observation sample for: ； Where X represents the instantaneous power value set, Indicates the corresponding working status code set, and the working status code starts from 0.

4. The method according to claim 1, characterized in that: The mixed probability distribution estimation model for: ； In the formula, is the latent feature vector of the observed sample; is the mixing probability corresponding to different working states; is the number of types of working status; Estimate the parameters of a model for a mixture probability distribution.

5. The method according to claim 1, characterized in that The objective function for: ； In the formula, Estimate parameters of models for mixture probability distributions; is the number of observed samples; and is the meta-parameter of the mixed probability model; the first term of the objective function is the working state identification loss term, is the cross entropy loss function, For the The working status soft label of each observation sample is calculated by the proportion of the number of corresponding working statuses in the observation sample. For the The mixed probability of different working states corresponding to observation samples; The second term of the objective function is the negative log-likelihood term, For the The latent feature vector of the observed samples, is the negative log-likelihood function, which contains the parameter estimation matrix of the mixed probability distribution. The mixed probability distribution is set to a mixed Gaussian distribution, and its parameters include the mixed probability matrix , expectation matrix , covariance matrix ; The third term of the objective function is the regularization term, is the covariance regularization function.

6. The method according to claim 1, characterized in that When the mixed probability distribution estimation model is trained end-to-end online using the observed samples and the objective function, it includes updating the parameters of the working state mixed probability distribution estimation model and updating the parameter estimation matrix of the mixed probability distribution.

7. The method according to claim 1, characterized in that The anomaly score for: ； In the formula, is the abnormal score of the current observation sample; is the estimated model parameters of the mixed probability distribution after training; is the instantaneous power value set of the current observation sample; Indicates that only the potential feature vector of the current observation sample is returned; They are the mixed probability matrix, expectation matrix, and covariance matrix after training; when the anomaly score is greater than the preset risk threshold It is determined that the energy storage power station sensor terminal is currently in an unsafe state.

8. A non-intrusive energy storage power station sensor terminal safety monitoring system, characterized in that: include: A stream data acquisition module is used to acquire the power consumption stream data of the energy storage power station sensor terminal using a non-intrusive power consumption measurement device; A stream data preprocessing module, used to extract observation samples from the stream data using a sliding window, wherein the observation samples include working state label codes and instantaneous power values ​​corresponding to a plurality of consecutive timestamps; Model building module, used to build a mixed probability distribution estimation model that describes the working status of the sensor terminal of the energy storage power station; An objective function construction module, used to construct an objective function of an end-to-end training mixed probability distribution estimation model, wherein the objective function includes a working state recognition loss term, a negative log-likelihood term, and a regular term; A model training module, used to perform end-to-end online training on the mixed probability distribution estimation model using the observed samples and the objective function to obtain a parameter estimation matrix of the mixed probability distribution; The online monitoring module is used to extract new observation samples from the stream data, and then calculate the anomaly score according to the trained mixed probability distribution estimation model and the parameter estimation matrix, so as to obtain the safety status judgment of the current energy storage power station sensor terminal.

9. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instruction is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images